Analytics and Knowledge

Management
 Data Analytics Applications
Series Editor: Jay Liebowitz

PUBLISHED
Actionable Intelligence for Healthcare
by Jay Liebowitz and Amanda Dawson
ISBN: 978-1-4987-6665-4
Analytics and Knowledge Management
by Suliman Hawamdeh and Hsia-Ching Chang
ISBN 978-1-1386-3026-0

Big Data Analytics in Cybersecurity

by Onur Savas and Julia Deng
ISBN: 978-1-4987-7212-9

Big Data and Analytics Applications in Government:

Current Practices and Future Opportunities
by Gregory Richards
ISBN: 978-1-4987-6434-6
Big Data in the Arts and Humanities: Theory and Practice
by Giovanni Schiuma and Daniela Carlucci
ISBN 978-1-4987-6585-5

Data Analytics Applications in Education

by Jan Vanthienen and Kristoff De Witte
ISBN: 978-1-4987-6927-3

Data Analytics Applications in Latin America and Emerging Economies

by Eduardo Rodriguez
ISBN: 978-1-4987-6276-2
Data Analytics for Smart Cities
by Amir Alavi and William G. Buttlar
ISBN 978-1-138-30877-0
Data-Driven Law: Data Analytics and the New Legal Services
by Edward J. Walters
ISBN 978-1-4987-6665-4

Intuition, Trust, and Analytics

by Jay Liebowitz, Joanna Paliszkiewicz, and Jerzy Gołuchowski
ISBN: 978-1-138-71912-5

Research Analytics: Boosting University Productivity and Competitiveness

through Scientometrics
by Francisco J. Cantú-Ortiz
ISBN: 978-1-4987-6126-0

Sport Business Analytics: Using Data to Increase Revenue and

Improve Operational Efficiency
by C. Keith Harrison and Scott Bukstein
ISBN: 978-1-4987-8542-6
Analytics and Knowledge
Management

Edited by
Suliman Hawamdeh
Hsia-Ching Chang
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2018 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-1386-3026-0 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmit-
ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
Contents

Preface ...............................................................................................................vii
Editors ................................................................................................................xi
Contributors .....................................................................................................xiii

1 Knowledge Management for Action-Oriented Analytics .......................1

JOHN S. EDWARDS AND EDUARDO RODRIGUEZ

2 Data Analytics Process: An Application Case on Predicting

Student Attrition ..................................................................................31
DURSUN DELEN

3 Transforming Knowledge Sharing in Twitter-Based Communities

Using Social Media Analytics...............................................................67
NICHOLAS EVANGELOPOULOS, SHADI SHAKERI,
AND ANDREA R. BENNETT

4 Data Analytics for Deriving Knowledge from User Feedback ...........121

KULJIT KAUR CHAHAL AND SALIL VISHNU KAPUR

5 Relating Big Data and Data Science to the Wider Concept

of Knowledge Management ................................................................141
HILLARY STARK AND SULIMAN HAWAMDEH

6 Fundamentals of Data Science for Future Data Scientists .................167

JIANGPING CHEN, BRENDA REYES AYALA, DUHA ALSMADI,
AND GUONAN WANG

7 Social Media Analytics .......................................................................195

MIYOUNG CHONG AND HSIA-CHING CHANG

8 Transactional Value Analytics in Organizational Development ........221

CHRISTIAN STARY

9 Data Visualization Practices and Principles.......................................251

JEONGHYUN KIM AND ERIC R. SCHULER

v
vi ◾ Contents

10 Analytics Using Machine Learning-Guided Simulations with

Application to Healthcare Scenarios ..................................................277
MAHMOUD ELBATTAH AND OWEN MOLLOY

11 Intangible Dynamics: Knowledge Assets in the Context

of Big Data and Business Intelligence ................................................325
G. SCOTT ERICKSON AND HELEN N. ROTHBERG

12 Analyzing Data and Words—Guiding Principles and Lessons

Learned...............................................................................................355
DENISE A. D. BEDFORD

13 Data Analytics for Cyber Threat Intelligence.................................... 407

HONGMEI CHI, ANGELA R. MARTIN, AND CAROL Y. SCARLETT

Index ...........................................................................................................433
Preface

The terms data analytics, Big Data, and data science have gained popularity in
recent years for a number of good reasons. The most obvious reason is the exponen-
tial growth in digital information and the challenge of managing large sets of data.
Big Data forms a challenge and opportunity at the same time. It is a challenge if
not managed properly and the organization does not make the needed investment
in the knowledge infrastructure recognizing the value of data as an organizational
asset. Knowledge infrastructure is made of several components including intel-
lectual capital (human capital, social capital, intellectual property, and content),
physical capital, and financial capital. What makes Big Data an opportunity is the
prospects of knowledge discovery from Big Data and the value of such knowledge
in enhancing an organization’s competitive advantage through improved products
and services, as well as enhanced decision-making processes.
Given the cost associated with managing Big Data, organizations must adopt
a knowledge management strategy in which Big Data is viewed as a key organi-
zational asset. This also includes making the necessary investment in data science
and data analytics tools and technologies. Knowledge management places a higher
emphasis on people and human capital as a key to realizing the concept of the
knowledge-based economy. This means any knowledge management strategy must
include a plan to educate and enhance the capacity of those working with Big Data
and knowledge discovery.
The shift toward the knowledge economy and the realization of the importance
of data as an organizational asset within the context of knowledge management
has given rise to the emerging fields of data science and data analytics. The White
House’s “Data to Knowledge to Action” initiative in 2013 aimed at building Big
Data partnerships with academia, industries, and public sectors. This initiative
led to the National Science Foundation (NSF) increasing the nation’s data science
capacity by investing in human capital and infrastructure development. The Big
Data to Knowledge (BD2K) initiative by the National Institutes of Health (NIH)
in 2012 and Google’s Knowledge Graph were also aimed at building big data
capacity. Such capabilities will be based on well-established knowledge infrastruc-
tures made of a network of individuals, organizations, routines, shared norms, and
practices. Building knowledge infrastructures requires human interactions through

vii
viii ◾ Preface

connecting people, organizations, and practices to facilitate knowledge discovery

from Big Data. However, relatively few organizations have developed data gover-
nance and knowledge management plans for handling and managing Big Data and
big data analytics projects.
Knowledge management is an interdisciplinary approach to dealing with all
aspects of knowledge processes and practices. Many of these activities are critical to
the notion of managing data and creating information and knowledge infrastruc-
ture in the long run. We all understand the importance of information organiza-
tion and data management for data (big or small) to be useful. The term “garbage
in, garbage out” is used to describe poor data and information organization prac-
tices. The long-term preservation of big data as part of the knowledge retention
process is crucial to the knowledge discovery process.
The process of transforming data into actionable knowledge is a complex pro-
cess that requires the use of powerful machines and advanced analytics techniques.
Analytics, on the other hand, is the examination, interpretation, and discovery of
meaningful patterns, trends, and knowledge from data and textual information. It
provides the basis for knowledge discovery and completes the cycle in which knowl-
edge management and knowledge utilization happen. This book examines the role
of analytics in knowledge management and the integration of Big Data theories,
methods, and techniques into the organizational knowledge management frame-
work. The peer-reviewed chapters included in this book provide an insight into
theories, models, techniques, applications, and case studies in the use of analytics
in organizations. The following are summaries of each chapter.
Chapter 1, “Knowledge Management for Action-Oriented Analytics” by
Edwards and Rodriguez traces the evolution of analytics, compares several practi-
cal concepts (analytics, business analytics, business intelligence, and Big Data),
and associates those concepts with knowledge management. They recommend
that organizations develop knowledge focuses on data quality, application domain,
selecting analytics techniques, and on how to take actions based on patterns and
insights derived from analytics. Such actions rest on leveraging people, processes,
and technology. They discuss the classifications of analytics from the dual per-
spectives of types of analytics techniques (descriptive, predictive, and prescriptive
analytics) and business management (strategic, managerial, operational, customer-
facing, or scientific analytics). From a global and cross-industry perspective, they
also provide examples that demonstrate how each individual analytics type corre-
sponds to one or multiple knowledge focuses.
Chapter 2, “Data Analytics Process: An Application Case on Predicting
Student Attrition” by Delen discusses a significant and coincidental resemblance
between the most popular data mining methodology, Cross-Industry Standard
Process for Data Mining (CRISP-DM), which addresses analytics processes, and
the Six Sigma-based DMAIC (define, measure, analyze, improve, and control)
methodology that not only improves organizational performances but supports
knowledge management processes. To exemplify the CRISP-DM approach, Delen
 Preface ◾ ix

introduces a specific case of data analytics in higher education regarding predicting

student success and student attrition.
Chapter 3, “Transforming Knowledge Sharing in Twitter-Based Communities
Using Social Media Analytics” by Evangelopoulos, Shakeri, and Bennett showcases
eight vignettes of social media analytics and illustrates how different dimensions
(such as time, location, topic, and opinion) can contribute to the knowledge base
of Twitter-based communities. Incorporating derived facts and derived dimen-
sions, they devise a knowledge base data warehouse (KBDW) schema for present-
ing transformed explicit knowledge in Twitter-based communities.
Chapter 4, “Data Analytics for Deriving Knowledge from User Feedback” by
Chahal and Kapur is a practice-oriented contribution. To make analytics of user
feedback data meaningful, they suggest connecting data analytics with data man-
agement and knowledge management as a three-step approach. They select the case
of the Indian government’s demonetization drive in 2016 and describe how the
government can better understand public perceptions across the country by means
of opinion mining and data mining of Twitter data.
Chapter 5, “Relating Big Data and Data Science to the Wider Concept of
Knowledge Management” by Stark and Hawamdeh outlines the concept of data
science, a core component being taught within master’s programs that focuses on
data as well as the skill set desired by current employers who seek to hire data scien-
tists. The chapter reviews some of the current data science and data analytics tools
being used to store, share, mine, model, and visualize Big Data. The chapter also
reviews some of the case study applications and their relevance to the concept of
managing data, small and big.
Chapter 6, “Fundamentals of Data Science for Future Data Scientists” by
Chen, Ayala, Alsmadi, and Wang considers data science as an interdisciplinary
field revolving around data and puts data science in the context of knowledge man-
agement processes. Through literature review and the analysis of 298 job postings,
they identify the desired knowledge and skills for data scientists. Additionally, they
reviewed existing data science programs in the United States and provided sugges-
tions for integrated curriculum design.
Chapter 7, “Social Media Analytics” by Chong and Chang starts with the evo-
lution of social media and analytics. Besides introducing the definitions and pro-
cesses of social media analytics from different perspectives, the chapter focuses on
identifying the main techniques and tools used for social media analytics.
Chapter 8, “Transactional Value Analytics in Organizational Development” by
Stary introduces an approach of value network analysis (VNA) and suggests how
organizations can use different proven methods (repertory grid, critical incident
technique, and storytelling) to elicit the tacit knowledge from stakeholder transac-
tions for value management. Stary interviewed 14 active analysts who had diverse
backgrounds and experiences in practical value management for international orga-
nizations in Germany and Austria. Those analysts appeared to reach the consensus
that the repertory grid technique combining qualitative with quantitative insights
x ◾ Preface

fits better in analyzing stakeholder transactions. In addition, Stary validated that

the repertory grid technique works effectively on constructing holomaps for collec-
tive sense-making during VNA.
Chapter 9, “Data Visualization Practices and Principles” by Kim and Schuler
demonstrates a hypothetical scenario: The Knowledge Management Corporation
(KMC) has been working with the fictional More Fun Toys company to assist in
understanding market trends and customer purchasing patterns over time. They
illustrate various data visualization tools that can support answering different types
of business questions to improve profits and customer relationships. The authors
showed that data visualization has a critical role in the advancement of modern data
analytics, particularly in the field of business intelligence and analytics.
Chapter 10, “Analytics Using Machine Learning-Guided Simulations with
Application to Healthcare Scenarios” by Elbattah and Molloy explores the poten-
tial opportunities of using data-driven machine learning to create new knowledge,
thus informing simulation models for better decision making. Like deep learning,
emerging machine learning techniques can greatly reduce human biases during
knowledge elicitation process. Elbattah and Molloy use the dataset of elder care
from Health Service Executive (HSE) of Ireland to depict an analytics use case
combining machine learning and simulation methods in elderly patient discharge
planning.
Chapter 11, “Intangible Dynamics: Knowledge Assets in the Context of
Big Data and Business Intelligence” by Erickson and Rothberg reframes the con-
ceptualization of intangible assets where the four-layer hierarchy framework starts
with data and information, moves from explicit knowledge to tacit knowledge, and
ultimately to insight and intelligence. In the proposed framework, they compare
and contrast the four layers in terms of definition, domain, source, exchange, range,
management, metrics, and indicates.
Chapter 12, “Analyzing Data and Words—Guiding Principles and Lessons
Learned” by Bedford exemplifies the approach to intelligent use of analytics and
thus enhancing knowledge management methods. The need for analysis to support
knowledge management goes beyond what has traditionally been available. The
chapter examines quantitative and qualitative analytics and their use including the
methods, tools, and technologies available. The technologies needed to do qualita-
tive and quantitative analysis should be built on a strong foundational understand-
ing of languages including the ability to interpret any content given to them at a
primitive language level.
Chapter 13, “Data Analytics for Cyber Threat Intelligence” by Chi, Scarlett,
and Martin provides an overview of data analysis uses in the field of cyber security
and cyber threat intelligence, including insider threat. The chapter examines the
process of collecting and organizing data as well as reviewing various tools for text
analysis and data analytics and discusses dealing with collections of large data-
sets and a great deal of diverse data types from legacy system to social networks
platforms.
Editors

Suliman Hawamdeh is a professor in the Department of Information Science

in the College of Information at the University of North Texas. He is the direc-
tor of the Information Science PhD program, one of the largest interdisciplinary
information science PhD programs in the country. He is the editor in chief of
the Journal of Information and Knowledge Management (JIKM) and the editor
of a book series on innovation and knowledge management published by World
Scientific. Dr. Hawamdeh founded and directed a number of academic programs
including the first Master of Science in Knowledge Management in Asia at the
School of Communication and Information at Nanyang Technological University
in Singapore. Dr. Hawamdeh has extensive industrial experience. He was the
Managing Director of ITC Information Technology Consultant Ltd, a company
that developed and marketed a line of software development products. He worked
as a consultant to several organizations including NEC, Institute of Southeast
Asian Studies, Petronas, and Shell. Dr. Hawamdeh has authored and edited sev-
eral books on knowledge management including Information and Knowledge Society
published by McGraw Hill and Knowledge Management: Cultivating the Knowledge
Professionals Published by Chandos Publishing, as well as a number of edited and
co-edited books published by World Scientific.

Hsia-Ching Chang is an assistant professor in the Department of Information

Science, College of Information at the University of North Texas. She received her
PhD in informatics and MS in information science from the University at Albany,
State University of New York as well as her MA in public policy from the National
Taipei University in Taiwan. Her research interests concentrate on data analyt-
ics, social media, cybersecurity, knowledge mapping, scientometrics, information
architecture, and information interaction.

xi
Contributors

Duha Alsmadi is a PhD candidate in information science at the College of

Information at the University of North Texas. She holds a master’s degree in
computer information systems. Her areas of interest include cloud computing,
e-learning, technology adoption, latent semantic analysis, as well as quantitative
research methods. She is currently working on her dissertation on the topic of
information sharing and storage behavior via cloud computing services.

Brenda Reyes Ayala is a PhD candidate in information science, College of

Information at the University of North Texas. She holds a master’s degree in
human–computer interaction. Her areas of interest include data science, informa-
tion retrieval, machine learning, and web preservation. She is currently working on
her dissertation on the topic of information quality in web archives.

Denise A. D. Bedford is currently an adjunct professor at Georgetown University’s

Communication Culture and Technology program, adjunct faculty at the Schulich
School of Business, York University, a visiting scholar at the University of Coventry,
and a Distinguished Practitioner and Virtual Fellow with the U.S. Department of
State. She teaches a range of graduate-level courses in knowledge management, enter-
prise architecture, and data sciences. Her current research interests include knowl-
edge architectures and knowledge engineering, knowledge economies and knowledge
cities, intellectual capital management, knowledge sharing behaviors, semantic anal-
ysis and text analytics, communities of practice, business architecture, document
engineering and content architectures, multilingual architecture, and search system
design and architectures. Dr. Bedford retired from the World Bank in 2010 where she
was the Senior Information Officer. In 2010, Dr. Bedford accepted the position of
Goodyear Professor of Knowledge Management at Kent State University. She retired
from this role in 2015. Dr. Bedford has also worked for Intel Corporation, NASA,
University of California Systemwide Administration, and Stanford University. Her
educational background includes a BA triple major in History, in Russian Language
and Literature, and in German Language and Literature from the University of
Michigan; an MA in Russian History also from the University of Michigan; an MS in
Librarianship from Western Michigan University, and a PhD in Information Science

xiii
xiv ◾ Contributors

from University of California, Berkeley. She is a certified enterprise architect. She

currently serves on several conference and journal editorial boards, and serves as an
associate editor of the Journal of Knowledge Management.

Andrea R. Bennett is a doctoral student in the Department of Marketing and

Logistics in the College of Business at the University of North Texas. She received
her master’s degree in public administration from the Institute of Public and
Nonprofit Studies at Georgia Southern University. Her research interests include
social media analytics, brand co-creation, and public-private partnerships.

Kuljit Kaur Chahal is an assistant professor in the Department of Computer

Science, Guru Nanak Dev University. Her research interests are in open source soft-
ware, distributed systems, data analytics, and machine learning. Chahal received
a PhD in computer science from Guru Nanak Dev University, Amritsar. She is a
POSSE graduate. Contact her at [email protected].

Hsia-Ching Chang is an assistant professor in the Department of Information

Science, College of Information at the University of North Texas. She received her
PhD in informatics and MS in information science from the University at Albany,
State University of New York as well as her MA in public policy from the National
Taipei University in Taiwan. Her research interests concentrate on data analytics, social
media, cybersecurity, knowledge mapping, scientometrics, information architecture,
and information interaction.

Jiangping Chen is a professor in information science in the College of Information

at the University of North Texas. She earned her PhD in information transfer from
the School of Information Studies at Syracuse University, New York. Dr. Chen
conducts research in data science, digital libraries, multilingual information access,
and information systems. She is the editor in chief of The Electronic Library, an
SSCI-indexed academic journal by Emerald Publishing Limited.

Hongmei Chi is an associate professor of Computer & Information Sciences at

the Florida A&M University, Tallahassee, Florida. She currently is the director of
the FAMU Center for Cyber Security and teaches graduate and undergraduate
courses in data mining and cyber security and conducts research in the fields of
Big Data and applied security. She has published articles related to data science,
parallel computing, and cyber security research and education. Her web page is
http://people.sc.fsu.edu/~hcc8471/.

Miyoung Chong is a doctoral candidate at the Department of Information Science

in the College of Information at the University of North Texas. Her research inter-
ests include open data, computational linguistics, new media theory, and media
analysis including social media, film, and mass media.
 Contributors ◾ xv

Dursun Delen is the holder of the William S. Spears Endowed Chair in

Business Administration and Patterson Foundation Endowed Chair in Health
Analytics, Director of Research for the Center for Health Systems Innovation,
and Professor of Management Science and Information Systems in the Spears
School of Business at Oklahoma State University (OSU). He received his PhD
in industrial engineering and management from OSU in 1997. He authored and
co-authored seven books and textbooks in the area of business analytics, deci-
sion support systems, business intelligence, data mining and text mining and his
research and teaching interests are in business analytics, data and text mining,
decision support systems, knowledge management, business intelligence, and
enterprise modeling.

John S. Edwards is emeritus professor and professor of knowledge management

at Aston Business School, Birmingham, UK. He holds MA and PhD degrees
from Cambridge University. His interest has always been in how people can
and do (or do not) use models and systems to help them do things. At present,
his principal research interests include how knowledge affects risk management;
investigating knowledge management strategy and its implementation; and the
synergy between knowledge management, analytics, and Big Data. He has writ-
ten more than 70 peer-reviewed research papers and three books on these top-
ics. He is a consulting editor of the journal Knowledge Management Research &
Practice.

Mahmoud Elbattah is a PhD candidate at the National University of Ireland,

Galway. He is a holder of the Hardiman Scholarship from the College of Engineering
and Informatics. His research interests are in the fields of simulation modeling and
machine learning. His e-mail address is [email protected].

G. Scott Erickson is a professor of Marketing in the School of Business at Ithaca

College, Ithaca, NY where he has also served as department chair and Interim
Associate Dean. He holds a PhD from Lehigh University, master’s degrees from
Thunderbird and Southern Methodist University, and a BA from Haverford
College. He has published widely on Big Data, intellectual capital, and business
analytics. He spent the 2016–2017 academic year studying knowledge networks
related to sustainability as a Fulbright National Science Foundation Arctic Scholar
at Akureyri University, Iceland. His latest book, New Methods in Marketing Research
and Analysis was published by Edward Elgar in 2017.

Nicholas Evangelopoulos is a professor of business analytics at the University of

North Texas, where he teaches business statistics, data mining, and Big Data ana-
lytics. He received his MS in computer science from the University of Kansas and
his PhD in business from Washington State University. His current research inter-
ests include text analytics, decision analysis, and applied statistics. His articles
xvi ◾ Contributors

appear in MIS Quarterly, Communications of the ACM, Decision Sciences, Decision

Support Systems, Communications in Statistics, and many others. His consulting
experience includes corporate projects in text analytics, predictive modeling, cus-
tomer attrition modeling, and policy evaluation. In 2013, he received the UNT
College of Business Senior Faculty Research Award, and in 2016 he was named
the 2016–2017 UNT College of Business Professional Development Institute
Fellow.

Suliman Hawamdeh is a professor and department chair of the Department of

Information Science in the College of Information at the University of North Texas.
He is the director of the Information Science PhD program, one of the largest inter-
disciplinary information science PhD programs in the country. He is the editor in
chief of the Journal of Information and Knowledge Management (JIKM) and the editor
of a book series on innovation and knowledge management published by World
Scientific. Dr. Hawamdeh founded and directed a number of academic programs
including the first Master of Science in Knowledge Management in Asia at the
School of Communication and Information at Nanyang Technological University
in Singapore. Dr. Hawamdeh has extensive industrial experience. He was the
Managing Director of ITC Information Technology Consultant Ltd, a company
that developed and marketed a line of software development products. He worked
as a consultant to several organizations including NEC, Institute of Southeast
Asian Studies, Petronas, and Shell. Dr. Hawamdeh has authored and edited several
books on knowledge management including Information and Knowledge Society
published by McGraw Hill and Knowledge Management: Cultivating the Knowledge
Professionals Published by Chandos Publishing, as well as a number of edited and
co-edited books published by World Scientific.

Salil Vishnu Kapur is currently pursuing a master’s degree in computer science

specializing in data science at Dalhousie University, Halifax, Nova Scotia. During
this research work, he was working with Capgemini in Pune, India as a senior ana-
lyst in the Big Data analytics domain. Kapur received a B.Tech in computer science
and engineering from Guru Nanak Dev University. In his undergraduate years,
he pursued research on service-oriented computing and published research in the
ICSOC and APSEC conferences. Contact him at [email protected].

Jeonghyun Kim is an associate professor at the Department of Information Science

in the College of Information at the University of North Texas. She received her PhD
from the School of Communication and Information at Rutgers University. Her
areas of research interest range from data curation, digital libraries, data visualization,
convergence issues surrounding libraries, archives, and museums to human computer
interaction. She has published numerous papers in journals, including the Journal
of the American Society for Information Science and Technology and the International
Journal of Digital Curation. She teaches “data visualization and communication”
 Contributors ◾ xvii

for the master’s degree in data science program at the Department of Information
Science in the College of Information at the University of North Texas.

Angela R. Martin is an intelligence specialist who works for New Generation Warfare
Study Group at USA TRADOC ARCIC as a Cyber Security Strategic Planning and
Development Analyst. She has 10 years of experience in the Department of Defense
in various roles to include human intelligence collector, senior interrogator, general
military intelligence analyst, human terrain specialist, research manager, knowledge
management and information technology supervisor, core curriculum instructor and
senior all source socio-cultural dynamics analyst.

Owen Molloy is a lecturer in information technology at the National University of

Ireland, Galway. His research interests are in the areas of business process model-
ing and simulation, software engineering, healthcare processes, and information
systems. His email address is [email protected].

Eduardo Rodriguez PhD, MSc., MBA Eduardo is the sentry endowed chair in
Business Analytics University of Wisconsin-Stevens Point. In his work he has created
the Analytics Stream of the MBA at University of Fredericton, Fredericton, Canada,
analytics adjunct professor at Telfer School of Management at Ottawa University,
Ottawa, Kansas, corporate faculty of the MSc in Analytics at Harrisburg University
of Science and Technology, Harrisburg, Pennsylvania, senior associate-faculty of the
Center for Dynamic Leadership Models in Global Business at http://nz.linkedin.com/
company/the-leadership-alliance-inc.?trk=ppro_cprof. The Leadership Alliance Inc.
Toronto Canada, and principal at IQAnalytics Inc. Research Centre and Consulting
Firm in Ottawa, Canada. He has been a visiting scholar Chongqing University,
Chongqing, China, and EAFIT University, Medellín, Colombia, for the Master of
Risk Management.
Eduardo has extensive experience in analytics, knowledge, and risk management
mainly in the insurance and banking industry. He has been the knowledge man-
agement advisor and quantitative analyst at Export Development Canada (EDC)
in Ottawa, regional director of PRMIA (Professional Risk Managers International
Association) in Ottawa, vice-president in Marketing and Planning for Insurance
Companies and Banks in Colombia. Moreover, he has a worked as a part-time pro-
fessor at Andes University and CESA in Colombia, author of six books in analytics,
and a reviewer of several journals and with publications in peer-reviewed journals
and conferences. He created and Chair the Analytics Think-Tank, organized and
Chair of the International Conference in Analytics ICAS, member of academic
committees for conferences in Knowledge Management and international lecturer
in the analytics field.
Eduardo holds a PhD from Aston Business School, Aston University,
Birmingham, UK, and MSc in Mathematics from Concordia University, Montreal,
Canada, Certification of the Advanced Management Program McGill University,
xviii ◾ Contributors

Montreal, Canada, and an MBA and bachelor in Mathematics from Los Andes
University, Bogotá, Colombia. His main research interest is in the field of Analytics
and Knowledge Management applied to Enterprise Risk Management.

Helen N. Rothberg is a professor of strategy in the School of Management at Marist

College, Poughkeepsie, NY. She holds a PhD and MPhil from City University
Graduate Center, and an MBA from Baruch College, City University of New
York. She is on the faculty of the Fuld-Gilad-Herring Academy of Competitive
Intelligence and is a principal of HNR Associates. She has published extensively on
topics including competitive intelligence and knowledge management. Her latest
book, The Perfect Mix: Everything I Know About Leadership I Learned as a Bartender
was published by Simon and Schuster in 2017.

Carol Y. Scarlett founded Axion Technologies, in January of 2016, to pro-

duce TRNG hardware based on experimentation she was performing. During
her seven years at Florida A&M University (FAMU), Tallahassee, Florida, she,
an associate professor in physics, has continued research in the areas of astron-
omy and particle physics. She has over 30 peer-reviewed publications and holds
3 patents on a novel technology for material identification and TRNG devices.
Through her work, she developed a technique to utilize magnetic birefringence
and microscopy to extend material identification to an unprecedented level of
sensitivity. As well, she holds two patents on methods to achieve enhanced optical
noise for purposes of encryption and simulations. She maintains active collabo-
rations with Brookhaven National Laboratory (BNL) and Lawrence Livermore
National Laboratory (LLNL).

Eric R. Schuler received his PhD in experimental psychology from the University
of North Texas, Denton, Texas. He has worked as a teaching fellow in the
Department of Psychology and as the research consultant for Department of
Information Science’s Information Research and Analysis Lab. His research inter-
ests include: refining best-practice quantitative techniques based on Monte Carlo
statistical simulations, measurement development and validation, and how belief
systems shift after a traumatic event. When Eric is not running R-code, he enjoys
playing Dungeons and Dragons.

Shadi Shakeri is a doctoral candidate in the Department of Information Science in

the College of Information at the University of North Texas. She received her MS
in library and information science from Kent State University. Prior to pursuing her
master’s degree, she worked in industry as a metadata specialist and system analyst in
Iran. Her research interests include data and network analytics, knowledge discovery,
knowledge management, data modeling, crisis informatics, and metadata designs.
 Contributors ◾ xix

Hillary Stark, with a background in business and marketing, focuses her research
towards nutrition marketing efforts, specifically the information relied upon by
individuals when purposefully making more healthful eating choices. She is excited
and encouraged by the present age of Big Data, as the tools made available to
analyze robust amounts of information, from conversations across social media
platforms to documents submitted on behalf of passing new dietary legislation,
can assist academics, health-care professionals, manufacturers, and the individu-
als themselves to be more in control of their actions through this newly acquired
knowledge. She is currently finishing her PhD studies in Information Science at the
University of North Texas, Denton, Texas, and enjoys running and volunteering in
her community in her free time.

Christian Stary received his diploma degree in computer science from the Vienna
University of Technology, Austria, in 1984; his PhD degree in usability engineer-
ing, and also his Habilitation degree from the Vienna University of Technology,
Austria, in 1988 and 1993, respectively. He is currently a full Professor of Business
Information Systems with the University of Linz. His current research interests
include the area of interactive distributed systems, with a strong focus on method-
driven learning and explication technologies for personal capacity building and
organizational development.

Guonan Wang is an adjunct professor in the Department of Information Science

in the College of Information at the University of North Texas. She earned her PhD
in physics from McMaster University.
Chapter 1

Knowledge Management
for Action-Oriented
Analytics
John S. Edwards and Eduardo Rodriguez

Contents
Introduction ..........................................................................................................2
Categorizing Analytics Projects ..............................................................................6
Classification by Technical Type ........................................................................6
Classification from a Business Perspective .........................................................7
How Analytics Developed .....................................................................................8
Analytics, Operations Research/Management Science,
and Business Intelligence..................................................................................9
Overview of Analytics Examples ..........................................................................10
Single-Project Examples ......................................................................................12
Strategic Analytics...........................................................................................12
Managerial Analytics.......................................................................................14
Operational Analytics .....................................................................................17
Customer-Facing Analytics .............................................................................18
Scientific Analytics..........................................................................................20
Multiple Project Examples ...................................................................................20

1
2 ◾ Analytics and Knowledge Management

Future Developments ..........................................................................................23

Organizational Environment ..........................................................................23
Political Environment .....................................................................................24
Analytics Workflow Embedded in Business Processes ......................................25
Conclusion ..........................................................................................................26
References ...........................................................................................................27

Introduction
Analytics, Big Data, and especially their combination as “Big Data analytics” or
“Big Data and analytics” (BDA) continue to be among the hottest current topics in
applied information systems. With the exception of artificial intelligence methods
such as those focused on deep learning, the emphasis on analytics is now moving
away from the purely technical aspects to the strategic, managerial, and longer-term
issues: not a moment too soon, some people would say.
In this chapter, we take a strategic perspective. How can an organization use
analytics to help it operate more successfully? Or even to help it become the orga-
nization that those who run it would like it to be? The need for more thought about
this is certainly evident, but the starting point is far from clear.
Our choice of starting point is to build on previous work on the relationship
between analytics, Big Data, and knowledge management. This is represented as an
action-oriented model connecting data, analytics techniques, knowledge, and pat-
terns and insights. These elements will be used to categorize and analyze examples of
analytics and big data projects, comparing the United Kingdom with other countries.
A particular focus is the difference between one-off projects, even if they lead to
ongoing results, such as sensor monitoring, and ongoing activities or a series of proj-
ects. In the latter, our model adds influences of knowledge on data (what to collect and
standardization of meaning), knowledge on techniques (new techniques developed in
the light of past examples), and action on knowledge (learning from skilled practice).
Links will also be made to the higher-level issues that drive analytics efforts - or at
least should do. These include the definition of problems, goals, and objectives, and
the measurement of organizational and project performance. This raises the question
of the relationship between knowledge management and strategic risk management.
Looking at the analytics “landscape,” we see that opinions, and even reported
facts, differ fundamentally about the extent to which organizations are already
using analytics. For example, several “white papers” attempt to describe the extent
and nature of the use of analytics by employing variants of the well-known five-
stage software engineering capability maturity model (CMM) (Paulk, Curtis,
Chrissis, & Weber, 1993).
A report commissioned from the UK magazine Computing by Sopra Steria
uses the same names as the CMM stages for respondents to describe their orga-
nization’s approach to data and analytics (Sopra Steria, 2016). The results were:
initial 18%, repeatable 18%, defined 19%, managed 33%, and optimized 12%.
 Knowledge Management for Action-Oriented Analytics ◾ 3

International Data Corporation (IDC) has a very similar five-stage model for
“BDA competency and maturity”—ad hoc, opportunistic, repeatable, manage-
able, and optimized—but their report gives no figures (Fearnley, 2015). MHR
Analytics (MHR Analytics, 2017), a UK consultancy group focusing on human
resource management, has developed a five-stage model for what they call the
“data journey,” comprising the following: unaware 33%, opportunistic 7%, stan-
dards led 18%, enterprise 19%, and transformational 30%. However, the fact that
these percentages add up to 107% does call their accuracy into question, even if
only in the proofreading of the report. Nevertheless, the impression given from
these three analyses is that at least half of the organizations responding are well on
the way to making good use of analytics and Big Data.
On the more skeptical side, the American Productivity and Quality Center
(APQC) reported in 2016 that four-fifths of organizations had not yet begun to
take advantage of Big Data (Sims, 2016). This does seem to be closer to our own
experience, although research by the McKinsey Global Institute (Henke et al.,
2016) found considerable differences between sectors, with retail more advanced
than most. As for the trend in Big Data and analytics use, again it may not be as
rapid as some articles—usually, those not quoting data—imply. A report by the
Economist Intelligence Unit (Moustakerski, 2015) suggests about a 10% movement
toward the strategic use of data between 2011 and 2015.
One reason for this difference of opinion/fact is the type of decisions for which
analytics and Big Data are being used. Are they the most significant decisions about
the strategic direction of the business, characterized by the Economist Intelligence
Unit in a report for PwC (Witchalls, 2014) as “big” decisions, or are they every-
day decisions? Later in the chapter we will use a classification by Chambers and
Dinsmore (2015) to help understand the effect of these differences.
The other crucial element is the extent to which the organization’s managers
accept, and plan for the use of, analytics; indeed, it might not be an exaggeration to
phrase that as the extent to which they believe in analytics. The Computing survey
(Sopra Steria, 2016) found that only 25% of respondents’ organizations had a well-
defined strategy and—not surprisingly in the light of that figure—that only 10%
of analytics projects were “always” driven by business outcomes, with 44% being
driven “most of the time.”
We see knowledge and its management as the connection between the orga-
nization and its strategy on the one hand, and the availability of data and the use
of analytics on the other. It is well-recognized that organizations with a culture of
knowledge sharing are more successful than those without such a culture (Argote,
2012). A key element in the use of BDA is the presence of a data-driven culture.
A data-driven, or data-oriented, culture refers to “a pattern of behaviors and prac-
tices by a group of people who share a belief that having, understanding and using
certain kinds of data and information plays a critical role in the success of their
organization” (Kiron, Ferguson, and Prentice, 2013, p.18). This is the link to the
“belief” in analytics that we mentioned earlier.
4 ◾ Analytics and Knowledge Management

Nevertheless, belief by itself is not enough for effective use of analytics. A bal-
ance needs to be struck between the “having” and “using” elements of the defini-
tion, and the “understanding” element. Spender (2007) has found that there are
three different types of understanding, although he prefers to call it organizational
knowing. These are data, meaning, and skilled practice. To use the distinction
identified by Polanyi (1966), some of this understanding and knowing is explicit
(codifiable in language), and some is tacit (not readily expressible). Crucially, for
skilled practice, the tacit knowledge dominates: even having good data and high-
quality technical analysis is not enough on its own.
Another element in our thinking is that the development of the relationship
between knowledge and analytics is in permanent evolution. Each time that an
analytics solution is obtained, it is then time to start the search for a new analyt-
ics solution based on the knowledge acquired from the previous analytics solution
implementation. Data changes, and that modifies the outcomes of the models, but
at the same time the models are changing, the problems are better defined, and the
scope is clearer, based on the experience in successive implementations. In addition,
there is a clear challenge in using the freshest data that it is possible to access. In
strategic risk, for example, every second the changes in the stock prices are poten-
tially modifying some results in models; in credit risk, there is a need to use a
permanent data flow that will potentially modify the outcomes for the decision-
making process (granting loans).
When we consider analytics, meaning and skilled practice refer both to the ana-
lytics and to the target domain where the analytics techniques are being applied.
Skilled practice and meaning also have a two-way relationship with knowledge;
they influence it and they are influenced by it.
This two-way interaction of meaning and skilled practice is illustrated by
the process of management control systems creation. In one direction the target
domain is the development of systems to assure the implementation of strategy in
organizations. The analytics domain concerns how to implement the knowledge
from accounting, cost control, and so on to develop performance indicators and
build up a measurement system for the organization’s performance at any level.
Nucor Corporation (Anthony and Govindarajan, 2007) is an example where the
organization started implementing knowledge management as a management con-
trol system, and evolved to develop capabilities to use data in its strategic develop-
ment and strategy implementation and in particular to the use of business analytics
(Hawley, 2016) for improving supply chain management practice, a core process
in the business. Thus a better understanding of the requirements of management
control systems led a transition from a general analytics view to specific analytics
application in a key area of Nucor Corporation’s strategy.
Figure 1.1 shows how knowledge, and the various elements involved in an ana-
lytics study, fit together when just a single study is considered. The knowledge falls
into one or more of four categories:
 Knowledge Management for Action-Oriented Analytics ◾ 5

Techniques
Data
(analytics)

Patterns/
Action
insights

Knowledge

Figure 1.1 The interactions between data, analytics, and human knowledge in a
single study. (From Edwards, J.S., and Rodriguez, E., Proc. Comp. Sci., 99, 36–49,
2016.)

◾ Knowledge about the domain, whether that is medical diagnosis, retail sales,
bank customer loyalty, etc.
◾ Knowledge about the data: are the sources reliable and accurate, are the
meanings clear and well defined, etc.
◾ Knowledge about the analytics techniques being used
◾ Knowledge about taking action: how to make real change happen; this can
be further subdivided into the aspects of people, processes, and technology
(Edwards, 2009)

We see analytics acting to supplement gut feel and intuition in the development
of solutions to business process problems. Knowledge management supports the
development of actions in organizations looking to reduce the storage of knowledge
that is never used. Knowledge that is not related to the process of finding solu-
tions is not adding value to the organization. The same is true of data: data with-
out analytics to create knowledge to be used in business solutions cannot provide
value. The Economist Intelligence Unit/PwC report mentioned earlier (Witchalls,
2014) indicates that several factors in big decisions can affect the use of analyt-
ics. Analytics requires preparation time to create a process that adds value in the
organization. Several decisions are made in a short time, forcing a decision based
on reactions. The necessary preparation time to avoid reactive solutions or solutions
and decisions enforced without good knowledge can be made available through a
systematic analytics monitoring system and continuous feedback regarding deci-
sions made and the value added by the analytics.
In the remainder of this chapter, we first offer some relevant definitions, then
look at the history of analytics, and move into the related issue of terminology.
This enables us to categorize the examples that we consider in the main body of the
chapter. We then conclude by considering likely future influences of the political
and business environments.
6 ◾ Analytics and Knowledge Management

Categorizing Analytics Projects

There is still no generally agreed definition of analytics, although data and models
feature in all those we have seen. For example, Davenport and Harris (2007, p.7)
define analytics as “the extensive use of data, statistical and quantitative analy-
sis, explanatory and predictive models, and fact-based management to drive
decisions and actions.” By contrast, the Institute for Operations Research and the
Management Sciences (INFORMS) explains that “analytics facilitates realization
of business objectives through reporting of data to analyze trends, creating pre-
dictive models for forecasting and optimizing business processes for enhanced
performance.” (Robinson, Levis, and Bennett, 2010).
We prefer those which are more action-oriented; that include a commitment to
action, or an expectation of it, rather than just a hope, or even no mention of action
at all, and so we use the Davenport and Harris definition.

Classification by Technical Type

Despite the lack of agreement on an overall definition, somewhat strangely there is
a much broader agreement about the different types of analytics. By far the most
commonly-seen is one that also comes (albeit rarely with proper credit) from the
work commissioned from Cap Gemini by INFORMS (Robinson et al., 2010),
which identifies three types: descriptive, predictive, and prescriptive analytics.
Descriptive analytics is “the use of data to figure out what happened in the past.
Descriptive analytics prepares and analyzes historical data and identifies patterns
from samples for reporting of trends” (Robinson et al., 2010). Specific techniques
under this heading include data modeling, visualization and regression analysis,
and newer techniques such as sentiment and affect analysis, web analytics, and
graph mining. Subsequently, Gartner Research extended this taxonomy (Chandler,
Hostmann, Rayner, & Herschel, 2011), restricting the category of descriptive ana-
lytics to the “what happened?” part, and separating that from the “why?”, which
they called diagnostic analytics, but this distinction does not seem to have caught
on more widely than in Gartner’s own materials.
Predictive analytics “uses data to find out what could happen in the future...
Predictive analytics predicts future probabilities and trends and finds relationships
in data not readily apparent with traditional analysis” (Robinson et al., 2010).
Specific techniques here include predictive statistical modeling, the whole range of
forecasting methods based on data, machine learning, and some forms of neural
network analysis.
Prescriptive analytics “uses data to prescribe the best course of action to increase
the chances of realizing the best outcome. Prescriptive analytics evaluates and deter-
mines new ways to operate, targets business objectives, and balances all constraints”
(Robinson et al., 2010). Techniques such as optimization and simulation are in this
group, along with others such as case-based reasoning, and neural networks that
 Knowledge Management for Action-Oriented Analytics ◾ 7

aim to recommend a particular course of action. Robinson et al. (2010) add that
“Most of us would probably agree that in fact most operations research (OR) tech-
niques reside in this space.”
Robinson et al. (2010) see the three types as a distinct hierarchy, commenting
that “Businesses, as they strive to become more analytically mature, have indicated
a goal to move up the analytics hierarchy to optimize their business or operational
processes. They see the prescriptive use of analytics as a differentiating factor for
their business that will allow them to break away from the competition.” This is
somewhat more contentious, much as its appeal to those pushing analytics services
and relevant IT products is obvious. INFORMS commissioned the 2010 work to
help decide how it addressed the rise of analytics as a hot topic in business. Haight
and Park (2015), in a report for Blue Hill Research about the Internet of Things
(IoT), also see descriptive, predictive, and prescriptive analytics as the stages in a
maturity model. In their case, the ultimate goal is complete automation, that is, to
“provide specific recommendations based on live streaming data to an employee or
to automatically initiate a process” (p.4).
To understand this issue better, we need to look at a classification of analytics
along a different dimension.

Classification from a Business Perspective

Chambers and Dinsmore (2015) identified five types of analytics viewed from the
perspective of the business: strategic, managerial, operational, customer-facing, and
scientific. We use this classification because of our interest in connecting analytics
knowledge with actions across the organization and across multiple projects.
Strategic analytics “typically provide high value and are operationalized through
an offline decision or process.” (Chambers and Dinsmore, 2015, p.66). These are
the key part of what the PwC report (Witchalls, 2014) calls the big decisions.
However, a big decision can have components of all five types. Consider the case of
a new product line development. The decision requires the strategic evaluation of
the business environment, and simultaneously the connection of the opportunity
with the capacity of the organization to take the opportunity and the acceptance
by, and expected value and impact on the stakeholders. The new product develop-
ment is at the same time a result of innovation that can be the outcome of scientific
analytics.
Managerial analytics, by contrast, “typically provide value through midterm
planning. The implementation of this type of analytics is often executed through
semi- or fully-automated processes.” (Chambers and Dinsmore, 2015, p.66).
Chambers and Dinsmore state that strategic analytics usually lead to action on
a timescale of one to three years, whereas managerial analytics are implemented in
three months to a year, but we think that this is a little misleading since timescales
vary so much between industry sectors. Nuclear power generation, for example, has
a vastly longer timescale than retail fashion.
8 ◾ Analytics and Knowledge Management

Operational analytics are “embedded into the front-line processes for an organi-
zation and are executed as part of the day-to-day operations. Operational analytics
ranges from real-time (now) to short-time horizons (today or this week).” (Chambers
and Dinsmore, 2015, p.66). No doubt about the timescales for this type.
Customer-facing analytics “typically provide value by providing insights
about customers. They also tend to range from real-time to short-time horizons.”
(Chambers and Dinsmore, 2015, p.67). The difference between this type and
the operational category is more clear cut in manufacturing than in some ser-
vice industry sectors. In the latter, sometimes all that matters is what the cus-
tomer thinks of the service.
Scientific analytics “add new knowledge—typically in the form of new intellec-
tual property—to an organization. The frequency may be periodic (every year) or
occasional (once every several years).” (Chambers and Dinsmore, 2015, p.67). This
type of analytics is specifically related to intellectual capital, and is rather different
from the managerial, operational, and customer-facing types, in that it is related to
building capacity or capability rather than to operations directly.
We will now begin to use these categories in considering the history and termi-
nology of analytics.

How Analytics Developed

Despite its recent “hot topic” status, analytics in organizations is far from a new
phenomenon (Holsapple, Lee-Post, and Pakath, 2014; Kirby and Capey, 1998). Its
roots can be traced back to at least the middle of the nineteenth century, with argu-
ably the best-known early examples being from public health rather than business
or industry. These include John Snow’s analysis of a cholera outbreak in London in
1854, and Florence Nightingale’s statistical work on causes of patient mortality in
British Army hospitals, also begun in the 1850s. Both of these examples consisted
of descriptive analytics, and both meet our action-oriented requirement. Snow’s
work led to the immediate action of removing the handle of a suspect street water
pump so that it could not be used, and in the longer term to confirming his theory
that cholera was transmitted through contaminated water. Nightingale’s work rev-
olutionized approaches to sanitation in the British Army (during the Crimean War,
poor sanitation was responsible for the death of more British soldiers than combat),
and subsequently she revolutionized British public health as well.
Reported examples of recognizable use of analytics in business or industry can
definitely be found from early in the twentieth century. We do not wish to spend
much space here debating whether or not Taylorism counts as analytics. However, it
is noteworthy that Kirby and Capey (1998), writing well before the recent upsurge
in interest in analytics, would surely have said that it wasn’t, describing Taylorism
as “grounded in a semi-intuitive feel for the practical aspects of manufacturing
processes rather than in the application of mathematical and statistical analysis”
 Knowledge Management for Action-Oriented Analytics ◾ 9

(Kirby and Capey, 1998, p.308). In the United Kingdom, there is the clear example
of the Shirley Research Institute, analyzing large amounts of data from the cot-
ton textiles industry in the 1920s (Kirby and Capey, 1998). In the United States,
management consultants such as Arthur D. Little, McKinsey & Company, and
Booz Allen Hamilton were well-known by the 1930s, and some of their work used
analytical approaches.
What we would now call analytics really began to gain traction in the 1950s,
spurred on by military uses in the 1940s. In the United Kingdom, J. Lyons & Co.
built the world’s first business computer in 1952 to carry out analytics work on food
sales in their chain of cafés. In the United States, UPS initiated its first corporate
analytics group in 1954 (Davenport and Dyché, 2013). For more early examples,
see Kirby and Capey (1998) for the United Kingdom and Holsapple, Lee-Post, and
Pakath (2014) for the United States.
Improvements in information technology in the twenty-first century have led to
the current boom in interest in analytics. Some authors have attempted to identify
different generations of analytics. Chen, Chiang, and Storey (2012) argued that
Business Intelligence and Analytics (BI&A), as they call it, evolved from BI&A
1.0 (database management system-based structured content) to BI&A 2.0 (web-
based unstructured content) and BI&A 3.0 (mobile and sensor-based content).
Davenport’s list, also of three generations, describes an evolution from Analytics
1.0 (the era of business intelligence) to Analytics 2.0 (the era of Big Data), and
moving toward Analytics 3.0 (the era of data-enriched offerings) (Davenport, 2013;
Davenport and Dyché, 2013). Davenport and Dyché (2013) actually date the three
eras, with Analytics 1.0 running to 2009, Analytics 2.0 from 2005 (so there is an
overlap) to 2012, and Analytics 3.0 from 2013.

Analytics, Operations Research/Management Science,

and Business Intelligence
As a result of this history of gradual evolution, there has been little standardiza-
tion of terminology, as those two lists of generations demonstrate. The reaction of
operations research/management science professionals to the growth of analytics
has split into two camps. Some have embraced the term as what they have been
doing all along. This can be seen in a widely-quoted example of how Coca-Cola
ensures orange juice products stick to the “product recipe” even though the char-
acteristics of the available ingredients vary (Stanford, 2013). This was publicized as
an analytics example, but the original article correctly describes it more specifically
as a technique; an algorithm (one of the many optimization techniques that come
under the prescriptive analytics category). It is a type of algorithm that has been
used in operations research/management science for decades. Only its complex-
ity and speed of operation differ from what could have been seen in the 1960s.
Indeed, one of this chapter’s authors worked on an application of a similar recipe
10 ◾ Analytics and Knowledge Management

algorithm for a brand of cat food more than 30 years ago. Again, this was prescrip-
tive analytics in action. Both this and the Coca-Cola orange juice example are
operational analytics, although in the cat food example the recalculation was only
done every few days. This was because of the purchasing cycle for the ingredients
rather than technological limitations; calculating the results (using what was then
called a mini-computer) took less than ten minutes.
On the other side of the divide, some operations research/management science
professionals seem to be ignoring the analytics “bandwagon” entirely. For example,
the study into improving paper production in Finland by Mezei, Brunelli, and
Carlsson (2016) uses detailed data and machine operators’ knowledge to fine-tune
the production machinery. This could appropriately have been described as an
example of predictive analytics, but neither the terms analytics nor Big Data appear
anywhere in their paper.
The problem with the bandwagon effect is that analytics has now reached the
point where consultancies and other players seek to differentiate themselves from
others in the market by using somewhat different terminology, and academics try to
make sense of the various labels. Early in the “boom” period, Davenport and Harris
(2007) described analytics as a subset of business intelligence. We have already seen
that Chen, Chiang, and Storey (2012) couple the two terms together. Business intel-
ligence, derived from the military sense of that term, and so meaning information
rather than cleverness, also has a long history. Holsapple, Lee-Post, and Pakath
(2014) point out that it was first coined at IBM in the 1950s (Luhn, 1958), although
it is often credited incorrectly to the Gartner Group in the 1990s (see e.g., Watson &
Wixom, 2007), much as Gartner helped to popularize the term. Summing up the
confusion, Watson & Wixom (2007) observed that “BI [Business Intelligence] is now
widely used, especially in the world of practice, to describe analytic applications.”
The MHR report (MHR Analytics, 2017) is a good illustration of how this
confusion of terminology continues. It has data analytics in its subtitle, but begins
with the statement “Business Intelligence (BI) is not a new topic,” and refers most
frequently to “BI and [data] analytics.” Just to add to the mix, the Computing report
(Sopra Steria, 2016) concentrates on the phrase “applied data and analytics.”

Overview of Analytics Examples

Rather than seeking to resolve these issues of detailed terminology, we will regard
any study or project that fits the definitions of descriptive, predictive, or prescriptive
analytics, as defined above, as being an example of analytics.
We will then look at the examples on two dimensions. One is the Chambers and
Dinsmore (2015) analytics categorization from the perspective of the business: stra-
tegic, managerial, operational, customer-facing, or scientific. The other is the focus
of the knowledge: domain, data, techniques, or action (Table 1.1). Each example
can fit only one analytics type, but may involve more than one knowledge focus.
Table 1.1 Examples Categorized by Analytics Type and Knowledge Focus
Knowledge Focus

Action (People,
Domain Data Techniques Process, Technology)

Analytics type Strategic Tesco portfolio UK banks Luxottica Uber

approach Hong Kong Efficency
U.S. Air Force Unit

Managerial Christie hospital Gelderse Vallei Warehouse Group Ford Motor

Caterpillar Marine hospital Schneider National Company
Warehouse Group Target UPS UPS
Marks & Spencer
Schneider National

Operational Ocado warehouse eBay eBay Coca-Cola

eHarmony Equifax
Geneia
Blue Cross-Nexidia

Customer- Ocado e-mail Bank of America SEB Tesco loyalty cards

facing HEQCO Dunnhumby contribution Ocado e-mail
to Tesco loyalty cards SEB
IPSoft contribution to SEB Colruyt

Scientific Panama Papers

Knowledge Management for Action-Oriented Analytics ◾
11
12 ◾ Analytics and Knowledge Management

To explain this further, let us look again at the two UK public health examples
from the nineteenth century. Although a “business” perspective does not strictly
apply, the categories of Chambers and Dinsmore can still be used. It is clear that
Snow’s first study was operational analytics, but later work became strategic analyt-
ics as more studies were carried out. In the first study, the crucial knowledge ele-
ments were knowledge of techniques and then knowledge of action. Snow needed
to develop a new technique—an early form of data visualization—to identify the
patterns in the data. Having identified the offending pump, Snow knew that just
putting a notice on it would not stop people using it, especially as many people in
London at that time could not read, and so the only way to stop them from drink-
ing the contaminated water was to disable the pump: the people element of action
was crucial. His later strategic work contributed to knowledge of the domain—the
theory of how diseases are transmitted.
Nightingale’s work also relied on knowledge of the people element of action,
specifically how to influence army generals and politicians to adopt her ideas, espe-
cially given that they were being proposed by a woman. Similarly, her work went on
to represent a major contribution to knowledge in the domain of public sanitation.

Single-Project Examples
Strategic Analytics
Let’s begin the section with another historical example. When retail banks in
the United Kingdom began to automate their processes in the late 1950s, based
on magnetic ink character recognition technology, the key piece of data was the
account number. As the use of computers expanded through the 1960s and 1970s,
databases continued to be structured with the account number as the primary key.
However, this was no help in identifying which accounts belonged to the same
customer. Integrating data to make the customer centrally involved considerable
rewriting of the banks’ systems, and only the twin drivers of concerns about Year
2000 (Y2K) problems and the rise of online banking finally persuaded UK banks
to make the necessary investment to do it properly. The whole process of switching
the data focus from the account to the person took about 20 years in the United
Kingdom, and arguably is still not entirely complete. This is an example of knowl-
edge about data linked to strategic analytics, which was simply not possible (even
for descriptive analytics) with the old account-centric computer systems.
For the next example, we present Tesco, the United Kingdom’s number one
supermarket, and one of the world’s largest retailers. Tesco has always been an
analytics leader among UK retailers, even before the term came into regular use.
For example, it pioneered the use of geographic information systems in United
Kingdom retailing in the 1980s, to analyze the “catchment area” of a supermarket,
aided by one valuable strategic insight. Tesco managers realized that deciding where
 Knowledge Management for Action-Oriented Analytics ◾ 13

to locate a new store and evaluating the performance of an existing store were basi-
cally variants of the same question: if there is a store located here, how well should
it perform? To gather the data and do the predictive analytics, they set up the Tesco
Site Research Unit. This enabled them to take a portfolio approach to opening new
stores (and acquiring stores from competitors) instead of considering each potential
new store decision in isolation. This was a strategic use of analytics resulting from
knowledge about the domain and was one of the elements that enabled Tesco to
become the United Kingdom’s number one supermarket, a position it has retained
ever since. We shall return to that latter point later in the chapter.
Another example is a project by the RAND Corporation (Chenoweth, Moore,
Cox, Mele, & Sollinger, 2012), developing supplier relationship management (SRM)
for the U.S. Air Force. Chenoweth et al. (2012) observed “As is the case with all mil-
itary services, the U.S. Air Force (USAF) is under pressure to reduce the costs of its
logistics operations while simultaneously improving their performance.” The USAF
addressed the problem on a premise of managing a good relationship with suppliers
based on the development of analytics capacity in the SRM system. The scope of
this project made it another strategic use of analytics based on domain knowledge.
The analytical capabilities were concentrated on the construction of appropriate
supplier scorecards that are associated with identification of improvement opportu-
nities in areas ranging from maintenance to resource planning systems. According
to Teradata, the main benefits for the USAF are concentrated on data and analytics
tools use “…to streamline workflows, increase efficiencies and productivity, track
and manage assets, and replace scheduled/time-based maintenance with condition-
based maintenance (CBM) – saving at least U.S.$1.5 million in one year.”
Returning to retail, the case of Luxottica Retail North America is also a strate-
gic view of analytics, this time based on knowledge of techniques. Luxottica sells
luxury and sports eyewear through multiple channels. The project consisted of the
integration of data and applications from internal and external sources. Its purpose
was to support the marketing strategy associated with data integration of multiple
distribution channels and the appropriate use of resources in marketing develop-
ment. According to IBM “…Luxottica gained a 360-degree view of its customers
and can now fine tune its marketing efforts to ensure customers are targeted with
products they actually want to buy.” (Pittman, 2016).
For an example of a focus on knowledge about technology leading to disruptive
innovation at the strategic level, we turn to Uber. The understanding of costs related
to taxi operations and the analysis of customers’ needs led to the development of a
substitute taxi service based on network and data analytics, where the technology
(phone applications) was crucial to effective action. Berger, Chen, and Frey (2017)
pointed out, “Unlike a traditional taxi business, Uber does not own any cars; instead,
it provides a matching platform for passengers and self-employed drivers and profits
by taking a cut from each ride.” The use of technology in understanding customers’
needs, transportation operation, and development is leading better customer services
and at the same time improvements in the organization’s performance.
14 ◾ Analytics and Knowledge Management

Innovation is not the only engine to connect strategic analytics and actions. The
Hong Kong Efficiency Unit (Hong Kong Efficiency Unit, 2017) illustrates strategic
analytics and all three aspects of knowledge about action: people, process, and
technology. To provide a good service to Hong Kong’s citizens, the Efficiency Unit
created a system for managing the answers to the contacts that people have with the
government. The use of analytics on data relating to prior contacts has provided a
capacity to anticipate complaints and offer a 24 × 7 service providing rapid answers
to the issues that have been reported.

Managerial Analytics
Moving to the slightly shorter timescales and lower-level decisions of managerial
analytics, we return to retail (remember we said it was one of the most advanced
sectors?), but cross to the other side of the globe and The Warehouse Group,
New Zealand’s largest retailer. Warehouse Stationery, part of the group, brought in
a count mechanism, which was based on thermal imaging, so they could understand
how many people were coming into a store. This was put alongside the transaction
data from the tills so that stores could understand how their conversion worked,
and higher-level managers could target stores that needed special emphasis. Kevin
Rowland, Group Business Intelligence Manager, commented “we very quickly
managed to create a series of dashboards that showed by hour, by day, to store level,
what was going on in terms of counts through the door and the conversion rate of
that… at the end of it, the GM of operations stood up and said, ‘I could kiss you,’”
(Ashok, 2014). In this case, the descriptive analytics project relied on knowledge
of techniques in data visualization, and in turn contributed to knowledge of the
domain-patterns of conversion rates in the stores.
Expanding our scope to a worldwide example, the Ford Motor Company oper-
ates the Global Warranty Measurement System. This is now available in the form
of an application, enabling Ford dealers to carry out descriptive analytics on war-
ranty repair costs, claims, and performance. It also enables Ford to gauge if dealers
will meet Ford’s expectations on their performance, so moving into the predictive
category. Cohen and Kotorov (2016) quote Jim Lollar, Ford’s Business Systems
Manager of Global Warranty Operations as saying: “We have always had data
on actual costs but we weren’t giving it to dealers” (p.107). The crucial knowl-
edge focus here was knowledge of action-process; how to deliver the data in a way
(an application) that was convenient to use. One of the most important contribu-
tions of the Global Warranty Measurement System is its effect on the dealerships’
performance management practice. Given the volume of data, the capabilities to
generate reports, and the governance of the data, it has been possible to develop
metrics that provide benchmarks as an indication of how the dealerships are operat-
ing. For example, it is possible to predict and manage the cost per vehicle serviced;
repairs per 1,000 vehicles serviced; and cost per repair (Cohen and Kotorov, 2016).
Dealerships can thus adjust their business processes to obtain better results.
 Knowledge Management for Action-Oriented Analytics ◾ 15

Returning to the United Kingdom, “the Christie” is a specialized cancer hospi-

tal in the National Health System (NHS) (the UK public healthcare system). The
hospital makes extensive use of statistical process control (SPC) in clinical processes
(McKenna, 2016b) to identify processes needing investigation or improvement.
SPC is not at all new as a technique for descriptive or predictive analytics, dating
back to the work of Shewhart in the 1920s, but the technology now available vastly
improves the speed of producing the control charts that originally had to be drawn
by hand. Areas of use at the Christie include monitoring patients’ length of stay,
bed availability, and discharge times, and in the future, it is planned to monitor
the cost of pharmacy dispensing. In all these examples we find the analytics work
contributing to knowledge of the domain.
Marks and Spencer (M&S) is another of the United Kingdom’s best-known
retailers, concentrating on clothing and food. They have used some form of descrip-
tive analytics to examine their sales for as long as current employees can remember.
Originally this was done informally in each store but later was centralized. Current
technology enables the analysis of much larger datasets, and lets the end users do it
themselves. This descriptive analytics leads to improved knowledge of the domain,
as illustrated by the following two examples from Goodwin (2014).
In one case, new clothing range launches were proving less successful than
hoped. Descriptive analytics revealed that the new clothes were available in stores
in the right total quantities but the wrong mixture of sizes. This is partly because
in the United Kingdom, as in many other countries, changes in diet mean that
people tend to be different in size and shape from their parents a generation ago.
This also led to supplier availability performance being shown as poor, because
people complained that a particular style was unavailable in their size, when in
fact the suppliers were delivering exactly what M&S had ordered, and on time.
The second example is from the other side of M&S’s business, food: smaller stores
were finding that they could not exploit new food products because the minimum
order they could place (determined by the tray size) was too large. So M&S halved
the tray size, and increased sales. Both examples show analytics as contributing to
knowledge about the domain.
Moving away from retail, our next example of managerial analytics is from
the logistics sector. Schneider National is one of North America’s largest truck-
ing firms, and is expanding the use of sensor data from its trucks. This includes
“a process where the sensor data, along with other factors, goes into a model that
predicts which drivers may be at greater risk of a safety incident. The use of predic-
tive analytics produces a score that initiates a pre-emptive conversation with the
driver and leads to less safety-related incidents.” (Davenport and Dyché, 2013).
Here, the input of knowledge about the techniques in use contributes to develop-
ing knowledge about the domain; for example, frequent hard braking is often an
indicator of a riskier driver.
Another logistics example comes from UPS, the world’s largest package deliv-
ery organization. They also have telematics sensors in their fleet of over 46,000
16 ◾ Analytics and Knowledge Management

vehicles (Davenport and Dyché, 2013). The data recorded from their trucks
includes, for example, their speed, direction, braking, and drive train perfor-
mance. UPS has embarked on a higher-level use of managerial analytics, which is
clearly OR on a big scale—predictive analytics. “The data is not only used to mon-
itor daily performance, but to drive a major redesign of UPS drivers’ route struc-
tures. This initiative, called On-Road Integrated Optimization and Navigation
(ORION), is arguably the world’s largest operations research project.” (Davenport
and Dyché, 2013, p.4) Davenport and Dyché (2013) also report that ORION had
“already led to savings in 2011 of more than 8.4 million gallons of fuel by cutting
85 million miles off of daily routes.” This is an example of using both knowledge
of suitable OR techniques, to take advantage of real online map data, and knowl-
edge of how to put them into action, the technology to carry out the calculations
quickly enough.
The next example uses somewhat similar technology—sensors in transportation—
but this time we see analytics that makes a contribution to knowledge of the domain.
It comes from Caterpillar Marine (Marr, 2017). Caterpillar’s Marine Division pro-
vides services to shipping fleet owners for whom fuel usage is a crucial factor affecting
profitability. One of Caterpillar’s customers knew that hull contamination (corro-
sion, barnacles, seaweed, etc.) must affect fuel consumption, but had previously had
no way of quantifying its effect. “Data collected from ship-board sensors as the fleet
performed maneuvers under a variety of circumstances and conditions—cleaned
and uncleaned—was used to identify the correlation between the amount of money
spent on cleaning, and performance improvements.” (Marr, 2017)
The outcome of the predictive analytics was the conclusion that ship hulls
should be cleaned much more frequently; roughly every six months as opposed
to every two years. Potential savings from these changes would amount to several
hundred thousand U.S. dollars per year (Marr, 2017).
We complete this section of managerial analytics examples with two exam-
ples where the knowledge focus is on the data, both related to healthcare. The
first is from the healthcare sector in the Netherlands (SAS Institute, 2016b).
A key performance indicator for surgery in public hospitals in the Netherlands,
as in the United Kingdom, is the length of time patients must wait before being
admitted for surgery. The Gelderse Vallei hospital’s waiting time for hernia sur-
gery had jumped. Rik Eding, Data Specialist and Information Analyst, explains.
“When we looked closer, it was because two patients postponed their operations
due to holidays. If we left these two cases out of consideration, our waiting period
had actually decreased” (SAS Institute, 2016b, p.13). Understanding outliers like
this in the data is a crucial element of descriptive analytics. Perhaps postpone-
ments by the patient should not be included in a calculation of the average wait-
ing time, but that would depend on the reason. A postponement because the
patient’s condition had worsened surely should be included. Standardization of
data is necessary for reliable analytics results, but it is not always easy to do it
appropriately.
 Knowledge Management for Action-Oriented Analytics ◾ 17

Wider issues still can arise. North American retailer Target offers a caution-
ary tale (Hill, 2012). They analyzed purchase histories of women who had entered
themselves on Target’s baby registry, and found patterns so clear that they could
estimate not just if a customer was pregnant, but roughly when her baby was due.
They went on to send coupons for baby items to customers according to their preg-
nancy scores. (It seems highly unlikely that this behavior would have been legal in
most European countries.) Unfortunately, they did not restrict this to customers on
the baby registry. This hit the press when an angry father went to the Target store
manager in Minneapolis to complain about the coupons that were being sent to his
daughter, who was still in high school. Now, the store manager had little under-
standing of the analytics process behind this marketing campaign, and was very
surprised to discover what was happening. It did transpire a few days later that the
girl was indeed pregnant. So, it was a technically sound use of predictive analytics
based on knowledge of data, but Target had not had the sense to restrict it only
to women already on the baby registry, and it earned them a great deal of dubious
publicity. Hence we have classified this under the managerial category. It is not
clear if Target still uses the pregnancy scores in its marketing.

Operational Analytics
The Coca-Cola orange juice case mentioned earlier is an example of operational
analytics, with the knowledge focus being on action—the technology needed to
carry out the predictive analytics fast enough.
Another brand name that is well-known worldwide gives us an operational ana-
lytics example where the knowledge focus is about the data and the techniques.
eBay uses machine learning to translate listings of items for sale into different lan-
guages, thus reaching more potential buyers (Burns, 2016). This is not a straight-
forward use of natural language processing (NLP), because the eBay listings that
form the data have specific characteristics. For example, does the English word
“apple” in a listing refer to the fruit (which should be translated) or the technology
company or brand (which should not)? Similarly, should an often-used acronym
like NIB, which stands for “new in box,” be translated or not? Some non-English
speakers will still recognize the meaning of NIB even if they do not know what the
acronym actually stands for. Therefore, eBay has had to develop its own statistical
algorithms to supplement more common NLP techniques, thus contributing to
knowledge of techniques (Burns, 2016). Machine learning for NLP does not fit as
well into the descriptive/predictive/prescriptive classification as most techniques,
but is most often regarded as predictive—the translation effectively serves as a fore-
cast of what the best translation might be.
Equifax, the credit scoring agency, has also developed its own techniques (the
knowledge focus) for key operational decisions (Hall, Phan, & Whitson, 2016).
Automated credit lending decisions have typically been based on logistic regres-
sion models. However, Equifax has now developed its own Neural Decision
18 ◾ Analytics and Knowledge Management

Technology (NDT), based on artificial neural networks with simple constraints,

which it is claimed delivers measurably more accurate results and produces “the
mandatory reason codes that explain the logic behind a credit lending decision”
(Hall et al., 2016). It also requires less human input, since “logistic regression
models often require segmentation schemes to achieve the best results” (Hall
et al., 2016). This is predictive analytics, but the dividing line between predic-
tive and prescriptive analytics often comes down simply to the use that managers
in the organization decide to make of the results: do they review them (which
makes it predictive), or always do what the software recommends (which would
be prescriptive)?
Geneia, a US healthcare organization, is another operational analytics example
where the principal knowledge focus is on techniques, supplemented by domain
knowledge (Hall et al., 2016). Geneia uses well-tried statistical and machine learn-
ing techniques, but has developed them using its own knowledge of the market and
regulatory privacy requirements. These comprise “statistical and machine learn-
ing techniques—including principal component analysis—to assimilate claims,
clinical, consumer, actuarial, demographic, and physiologic data…[b]inary logistic
regression, a straightforward classification technique, helps identify predictors and
develop risk-scoring algorithms to identify where sick patients fall on a wellness
spectrum and the most appropriate remediation” (Hall et al., 2016, p.20).
That example related to relatively traditional databases and analytics methods.
However, even in healthcare the use of nonstructured data for providing or improv-
ing services is expanding. Blue Cross reported an application of speech recogni-
tion for improving patients’ experience. They use Nexidia solutions, so clearly the
knowledge focus is on techniques. The purpose has been the development of an
effective and efficient voice of the customer (VoC) by using data available from the
contacts and interactions between customers and the company with the purpose
of improving customer service. The analytics system allows not only a better quan-
tification of the issues to solve based on the data collected but also supported the
solution of problems, opening the coordination of areas and employees involved in
a customer’s solution. This coordination allows a reduction of contact with employ-
ees and provides better service (Nexidia, 2014). This example also shows how the
analytics type and knowledge focus can differ even when the broad application area
is the same; contrast this Blue Cross case with the Hong Kong Efficiency Unit case
above, and the Ocado case in the next section.

Customer-Facing Analytics
Gaining insights into customers and their behavior is one of the most active areas
of analytics at present, perhaps because of the perception that this has led to the
success of companies like Amazon.
For our first example in this section, we look to Tesco again. In the 1990s, Tesco
was the first UK retailer to make a success of using customer data from loyalty
 Knowledge Management for Action-Oriented Analytics ◾ 19

cards, in collaboration with market intelligence company Dunnhumby. Tesco had

to issue its own cards, because UK data protection laws, unlike those of some other
countries, do not permit retailers to identify customers from their credit card data.
The crucial knowledge focus here for Tesco was on action—the people aspect of
knowing what benefits (discount coupons, etc.) customers needed to make it worth
their while to obtain a loyalty card and remember to use it. This illustrates our point
about action-orientation very well. Without these insights, the predictive analytics
(knowledge about techniques from Dunnhumby) would not have achieved any-
thing for Tesco.
One of Tesco’s UK competitors is Ocado. Ocado has no physical stores. As an
online-only grocer, most of their customer contact is via email. Coping with the
volume of emails can be a particular problem at the holiday season or during freak
weather conditions that delay deliveries. Ocado is now trialing machine learning to
help respond to the emails it receives. There are two success criteria for this use of
predictive analytics: “does the model behave in the same way a human would?” and
“is it helping to speed up resolution times?” (Donnelly, 2016). It is evident that the
first of these criteria relates to knowledge of the domain, and the second to action-
oriented knowledge about the process.
Tesco still uses its loyalty cards, but these days delivers most of the benefits to
customers online, including discount coupons. However, Colruyt, one of the larg-
est supermarket chains in Belgium, combines old and new technology. Colruyt still
sends out paper coupons for discounts, but by using predictive analytics, Colruyt
can “calculate purchasing probabilities based on past customer behavior, as well
as on household and demographic information stored in its database. Of the 400
products on promotion during a given two-week period, each household receives
four pages of coupons specifically tailored to it, down from 32 pages previously.”
(Alford, 2016). Again, this revolves around knowledge of action, both people (the
customers and how they behave) and the processes (obviously Colruyt’s consumers
are still happy with paper coupons).
Another European country gives us a very different example of predictive ana-
lytics. Skandinaviska Enskilda Banken (SEB) is a large full-service Swedish bank.
Toward the end of 2016, SEB introduced Amelia to its customer service operations.
Amelia is machine learning software; a “chatbot” that can understand text and
respond to customer queries (Flinders, 2016). Knowledge about action is again
central here, with a people focus. Flinders (2016) describes the key issues as what
people will put up with, and different people wanting different things, since the
main channel for customer queries at present is the phone, rather than text-based
chat. Domain knowledge may turn out to be relevant (as in the eBay case earlier),
and also the knowledge of language processing techniques contributed by IPSoft,
for whom SEB represents the first users of Amelia in a language other than English.
In a different sector, student retention and attrition are topics of great concern
to education authorities. Domain knowledge of the causes and factors affecting the
levels of final graduation indicators is a very important area of analytics for them.
20 ◾ Analytics and Knowledge Management

Lopez-Rabson and McCloy (2013) carried out a descriptive analytics study of col-
leges in Toronto, Canada, for the Higher Education Quality Council of Ontario.
Their study discovered factors that provide insights to education institutions to
develop strategies and tactics to support students’ success, so it is indeed customer-
facing even though it is in the strategic knowledge category.
Closing this section, we look at Bank of America for an example of predictive
analytics where knowledge of the data is the main driver (Davenport and Dyché,
2013), although the business goal, as with all customer-facing analytics, is under-
standing the customer. The developments come from combining customer data
across all channels. The predictive models establish that “primary relationship cus-
tomers may have a credit card, or a mortgage loan that could benefit from refinanc-
ing at a competitor. When the customer comes online, calls a call center, or visits a
branch, that information is available to the online app, or the sales associate to pres-
ent the offer” (Davenport and Dyché, 2013, p.16) There is also a minor action focus
on process knowledge here, in that having combined the data from all channels, the
offer is also made available across all channels, and an incomplete application (e.g.,
online) can be followed up through a different channel (e.g., an e-mail offering to
set up a face-to-face appointment).

Scientific Analytics
Our chapter devotes less space to scientific analytics than the other categories, since
by definition, the purpose of work in this category is to add to an organization’s
intellectual capital. The close links between intellectual capital and knowledge
management should mean that the case for the relevance of knowledge to action
does not need to be made. Nevertheless, one example is particularly instructive. The
so-called Panama Papers (McKenna, 2016a) represented a success for descriptive
analytics using the new technique of graph databases. The papers comprised a huge
set of documents (11.5 million) from the Panama-based legal services firm Mossack
Fonseca, which specializes in offshore organizations, including those in countries
often regarded as tax havens. The new technique made it possible for news orga-
nizations such as the International Consortium of Investigative Journalists (ICIJ),
the BBC, UK newspaper the Guardian, and German newspaper the Süddeutsche
Zeitung to analyze connections in the data in ways that would have been near-
impossible before. So the knowledge of the technique turned the Panama Papers
into a resource they would not otherwise have been, especially for the ICIJ.

Multiple Project Examples

It is not difficult to find many other examples of successful analytics use online in
blogs and press releases. Documented successful application is far less common in
the academic literature, with the typical 2–3 year time lag between data collection
 Knowledge Management for Action-Oriented Analytics ◾ 21

Techniques
Data
(analytics)

Patterns/
Action
insights

Knowledge

Figure 1.2 The interactions between data, analytics, and human knowledge
over several studies. (From Edwards, J.S., and Rodriguez, E., Proc. Comp. Sci., 99,
36–49, 2016.)

and publication. However, even in blogs and press releases, relatively few articles
look at the evolution of analytics in an organization, and the way that the orga-
nization progresses from one project to another. We have already hinted at some
examples of this, and in this section we consider them in more detail.
The essential relationships between data, analytics techniques, patterns/insights,
knowledge, and action remain the same across a series of projects or studies, but
we expect to see three additional influences, as shown by the dashed arrows in
Figure 1.2. These are all types of learning: influences of knowledge on data (what to
collect and standardization of meaning); of knowledge on techniques (new analyt-
ics techniques developed in the light of past examples); and of action on knowledge
(learning from skilled practice).
Beginning again with Tesco, many of the analytical activities are still the same
as when they first went into analytics decades ago, like sales forecasting. But there,
for example, their priority is now real-time analytics: doing the same things, but
doing them much faster (Marr, 2016). This involves both newer technology, like
Teradata, Hadoop, and a data lake model, and also changing business processes. So
we see the contribution of knowledge of action—processes and technology—with
Tesco learning a very great deal from their existing skilled practice, and trying to
preserve their lead over competitors in this respect. The actual examples can be
managerial, operational, or customer-facing, depending on the detail. However,
especially at the managerial level, at present it can take 9–10 months to progress
from data to action (Marr, 2016). This is not unusual: the findings of Halper (2015)
were that just the stage of putting an agreed predictive model into action can take
several months. She found that only 31% of respondents said this process took
a month or less, with 14% responding nine months or more, and the mean and
modal response being three to five months.
Other examples of learning from skilled practice include Ocado and SEB. Before
the customer email project, Ocado already used machine learning in its warehouse
operations, co-locating items that are frequently bought together to reduce packing
22 ◾ Analytics and Knowledge Management

and shipping time (Donnelly, 2016). Similarly, SEB first used Amelia to answer que-
ries coming to its internal IT help desk (Flinders, 2016) before rolling out their
customer-facing system. Both of these were operational examples of predictive ana-
lytics, whereas the later projects were customer-facing. We thus see that the analytics
type and the knowledge focus can change between related studies or projects, without
affecting the possibility of transferring learning—of action influencing knowledge.
Another change of type is the plan for UPS’s ORION to include online map and
traffic data, so that it will reconfigure a driver’s pickups and drop offs in real time,
moving from managerial to operational (Davenport and Dyché, 2013). However, as
far as we know, this part of the system has not yet been rolled out.
In both the Warehouse Group and Marks & Spencer, the new knowledge cov-
ers the whole domain of using analytics. Kevin Rowland of the Warehouse Group
again: “I would actually design something in the beginning and give it to the busi-
ness, and ask them what they think, as opposed to going to the business with
a blank page and asking them what they want. Now they understand what the
tool can do but back then they didn’t.” (Ashok, 2014). Equally, Paul Williams,
M&S Head of Enterprise Analytics, observes: “People are asking questions they
wouldn’t have asked before. And asking one question breeds curiosity, often leading
to another 10, 20, or 30 questions.” (TIBCO Software, 2015).
In terms of learning about the data, the UK banking example has relied crucially
on standardization of meaning, by shifting from account number to a customer
identifier, even when that leads to counter-intuitive structures (in database terms)
such as two “unique” customer identifiers for each joint account. As for Caterpillar
Marine, James Stascavage, Intelligence Technology Manager, concluded: “I think
the best lesson we learned is that you can’t collect too much information…the days
of data storage being expensive are gone…if you don’t collect the data, keep the
data and analyse the data, you might not ever find the relationships.” (Marr, 2017)
Equifax is a prime example of developing new techniques in the light of past
examples. The NDT’s increased accuracy could also lead to credit lending in a
broader portion of the market, such as new-to-credit consumers, than previously
possible (Hall et al., 2016).
The history of the online dating website eHarmony shows the evolution of
data knowledge and the operational analytics solutions that were developed
from it. eHarmony started at the end of the last century with the original use of
questionnaire-style variables and data to predict matching and compatibility for
good relationships among couples. Given the need to develop trust among the
service users, eHarmony evolved to use social media data to manage (some of)
the dating uncertainty. Nevertheless, the business these days is affected not only
by the customers’ use but also by the business model. As Piskorski, Halaburda,
and Smith (2008) pointed out “eHarmony’s CEO needs to decide how to react to
imitations of its business model, encroachment by competing models and ascen-
dance of free substitutes.” Thus, as we said earlier, all five types of analytics may
turn out to be linked.
 Knowledge Management for Action-Oriented Analytics ◾ 23

Future Developments
In this section we present three main aspects of analytics and knowledge in the
action-oriented view: organization environment evolution, political environment
evolution, and transforming the analytics workflow into models that can be embed-
ded into business processes and evolve according to new data input.

Organizational Environment
One of the key difficulties in organizational innovation revealed by the busi-
ness process re-engineering movement in the 1990s (Davenport and Short, 1990;
Hammer, 1990) was the presence of organizational silos: vertical divisions that
acted as barriers to better organizational communication and hence performance.
Over 25 years later, silos still rule in far too many organizations. Henke et al.
(2016) point out that the old organizational silos that often still exist have been
joined by data silos: different systems that cannot be connected to each other.
They give examples of the barriers presented by these silos in three sectors: bank-
ing, manufacturing, and smaller specialized retailers. A survey carried out by Cap
Gemini Consulting in November 2014 (reported by Burns, 2015, p.11) backs this
up. They found that “data scattered in silos across various units,” cited by 46% of
respondents, was the greatest challenge in making better use of Big Data.
Virgo (2017) understands these issues better than most, because of his particu-
lar interest in IT in national and local government, observing that there seems to
have been little improvement generally in the past 10 years. He comments on a
successful example of smart city analytics from New York (unfortunately his article
gives no more details) that “the breakthrough had been to get departments to share
sensitive information on problems they were shy of discussing with others.” Virgo
goes on to identify the three main reasons for this shyness as:

1. Fear that others will learn just how poor (quality, accuracy, etc.) our own
data is
2. The inability to agree on common terminology
3. Fear that others will abuse “our” data

These are specific examples of what Alvesson and Spicer (2012) have labeled a
stupidity-based theory of organizations.
Organizational silos come under the people heading of action issues, and data
silos under the technology (and people) headings. However, taking action to move
forward can also raise difficulties of changing processes. In Equifax, the company’s
operations are entirely structured around the existing analytics techniques. This is
an instance of the well-known problem that first emerged in the 1970s: it is usually
easier to computerize a manual process than to switch from one computer system
to another.
24 ◾ Analytics and Knowledge Management

The last big silo, especially for customer-facing analytics, is the one surround-
ing personal data. How much of our personal data are we willing to share with
organizations? A report for the Future Foundation/SAS about UK attitudes (SAS
Institute, 2016a) finds that younger people are much more willing to share personal
data than older people when there is some direct benefit in it for them, but less
willing when the benefits are for society as a whole (e.g., sharing energy use data to
reduce the risk of brownouts or blackouts).
Despite these barriers, some organizations have succeeded in analytics and will
endeavor to continue to do so. Marr (2016) reports four key challenges for Tesco in
the future. Below we relate them to our discussion:

◾ Continuing to gain a better understanding of the changing nature of con-

sumer behavior. This is learning from skilled practice.
◾ Creating efficiencies in their logistics and distribution chains, to keep down
costs and minimize environmental impact. This is also learning, but with
more emphasis on new techniques.
◾ Facing up to the challenge of emerging business models that compete with
their own (such as Ocado, for example). Tesco’s business model is still based
around their network of stores despite a large and successful online pres-
ence: Tesco is regularly in the top four UK online retailers across all sectors.
This is clearly the main strategic challenge, but Tesco has used analytics to
help with strategy before, as we saw in their portfolio approach to new store
development.
◾ Reducing the amount of food that goes to waste at their stores. This is a new
direction for Tesco’s operational analytics.

Political Environment
As the use of analytics, especially predictive and prescriptive systems based on
deep machine learning, becomes more widespread, the regulatory and gover-
nance environment will undoubtedly need to change to cope with it. This is not
a new issue. The practice of “redlining” in the banking and insurance industries,
whereby loans or insurance are refused to anyone in a particular geographic area
(originally shown by a red line on a map) has been outlawed by many laws, legal
rulings, and judgments over the past 50 years. Similarly, the European Court of
Justice ruled in 2011 that it was against European Union equality laws to use gen-
der in calculating car driver insurance premiums. A machine learning system that
“reinvented” redlining or gender discrimination would put its owners in a difficult
legal position.
As a consequence, the UK Labour Party (currently the main opposition) is call-
ing for the algorithms used in big data analytics to be subject to regulation (Burton,
2016). This would not be a long step from current policy: the UK National Audit
 Knowledge Management for Action-Oriented Analytics ◾ 25

Office already has a detailed framework for the evaluation of models used in gov-
ernment departments (White & Jordan, 2016). However, enforcement in the pri-
vate sector would present a new challenge.
With a worldwide perspective, many feel that artificial intelligence (AI) needs
regulation, resulting in developments such as the Asilomar AI Principles (Future
of Life Institute, 2017), endorsed by Elon Musk and Stephen Hawking amongst
others. Two of the principles are very relevant to our discussions:

#7 Failure Transparency: If an AI system causes harm, it should be possible to

ascertain why.
#8 Judicial Transparency: Any involvement by an autonomous system in judi-
cial decision-making should provide a satisfactory explanation auditable by
a competent human authority.

Judicial use of analytics would certainly be a big step forward from credit scoring,
a chatbot, or even a self-driving car.

Analytics Workflow Embedded in Business Processes

In the search to improve knowledge management for action-oriented analytics, one
factor that is crucial is to keep updating the results based on the permanent feed
of data used. The new data and the changes in the outcomes will lead to a better
design of intelligent systems in analytics. Big Data leads the need to create adap-
tive systems given the potential of creating new knowledge, learning, and modi-
fying actions in an ongoing changing business environment, as shown above in
Figure 1.2. The approach to the analytics process is a continuous set of predefined
and sequential steps that requires adaptive development according to the ongoing
feedback (Rodriguez, 2017). Adaptive means that the analytics system is in a per-
manent state of reviewing the feedback that new data, tests, and actions produce.
The tools that the analytic system possesses to create Predictive Model Markup
Language (PMML) (Pechter, 2009) outcomes are a key component of the analyt-
ics system to generate actions in business processes. This means that, for example,
predictive analytics at any level (strategic, managerial, etc.) will have a great deal
to contribute when new data arrive, and the systems will be able to recalculate
and develop new outcomes to manage the decision-making process. Adaptive mod-
els will require organizational changes because what is true now (belief) will not
necessarily be true tomorrow. In risk management, it is crucial to understand the
variations that are expected, in particular, strategic risk and credit risk. In strategic
risk, opportunities can be important today because of diverse factors that change
permanently and modify the expected worth of big decisions, while credit risk can
have new variables in the profiles to define good or bad prospects, thus modifying
managerial decisions.
26 ◾ Analytics and Knowledge Management

Conclusion
We have explained why we believe that knowledge management is the key to suc-
cessfully putting analytics into action. We discussed many examples, looking at
data, the analytics techniques, and the people, process, and technology issues of
implementing the system.
Despite barriers such as silos, and the potential legal issues of relying too much
on a machine learning system that cannot explain its decisions in terms that a
human can understand, the use of analytics will surely continue to expand rapidly.
The millennial generation seems to be very much at home with it.
However, at the heart of any analytics project there has to be data. Sometimes
the data will pose problems that cannot be overcome. For example, predicting how
long it will take to develop computer software has been an issue for over 40 years.
Sadly, the data required to drive a predictive model giving an accurate (within
10%) forecast is simply not available at the time when the estimate is needed for
business purposes (Edwards and Moores, 1994). Alternatively, the data may not be
collected, even now. As we saw, Hall et al. (2016) reported Equifax as claiming that
the new technique delivers “measurably more accurate” results than their older one.
This begs the question of what data, if any, they had on loans that were declined.
It is almost a logical impossibility to know that a loan that was declined should in
fact have been offered.
Thus an increasing strategic risk stems from poor understanding of what to do
with the knowledge created from data. Knowledge creation through analytics can
be costly in both the search for solutions and in their implementation. Companies
can lose this investment because they are insufficiently prepared to use the new
knowledge (in our terms, lack of domain knowledge). Another strategic risk is to
consider data as a resource and not as an asset. If data are an asset, adequate gover-
nance and care should be provided. The beauty of data and analytics development
is that in many cases organizations will be able to do more with the same resources.
Scope economies will take a predominant place in benefits generation instead of
continuing only with the paradigm of scale economies and growth. Several com-
panies are returning to their core business to use their resources in a better way to
create wealth (GE is one example); data are a particular resource to use to manage
uncertainty and risk.
Turning to the strategic risks of data analytics, these are mainly associated with
the creation of an analytics process that can support accuracy in predictions and
prescriptions, and possibly most important the generation of meaning and under-
standing for the organization’s management. And of course, it all depends on the
accuracy of the data…
So, even in the age of Big Data, analytics, and deep machine learning, an adage
that is as old as the use of computers holds true—“garbage in, garbage out”!
 Knowledge Management for Action-Oriented Analytics ◾ 27

References
Alford, J. (2016). Keep them coming back: Your guide to building customer loyalty with
analytics. Cary, NC: SAS Institute.
Alvesson, M., & Spicer, A. (2012). A stupidity-based theory of organizations. Journal of
Management Studies, 49(7), 1194–1220. doi:10.1111/j.1467-6486.2012.01072.x.
Anthony, R., & Govindarajan, V. (2007). Management control systems. New York, NY:
McGraw-Hill.
Argote, L. (2012). Organizational learning: Creating, retaining and transferring knowledge.
New York, NY: Springer Science and Business Media.
Ashok, S. M. (2014). Case study: Warehouse Group does more with data. Computerworld
New Zealand. Retrieved January 30, 2018 from http://www.computerworld.co.nz/
article/541672/case_study_warehouse_group_does_more_data/
Berger, T., Chen, C., & Frey, C. B. (January 23, 2017). Drivers of disruption? Estimating
the Uber effect. Retrieved March, 11, 2017, from http://www.oxfordmartin.ox.ac.uk/
downloads/academic/Uber_Drivers_of_Disruption.pdf
Burns, E. (2015). Big data challenges include what info to use—and what not to.
The Key to Maximizing the Value of Data (pp. 7–16): Information Builders/
SearchBusinessAnalytics. Retrieved January 30, 2018 from http://docs.media.bitpipe.
com/io_12x/io_122156/item_1108017/InformationBuilders_sBusinessAnalytics_
IO%23122156_Eguide_032315_LI%231108017.pdf.
Burns, E. (2016). Ebay uses machine learning techniques to translate listings. Retrieved
January 30, 2018 from http://searchbusinessanalytics.techtarget.com/feature/EBay-
uses-machine-learning-techniques-to-translate-listings?utm_content=control&utm_
medium=EM&asrc=EM_ERU_72462481&utm_campaign=20170210_ERU%
20Transmission%20for%2002/10/2017%20(UserUniverse:%202298243)&utm_
source=ERU&src=5608542
Burton, G. (2016). Labour Party to call for regulation of technology companies’ algo-
rithms. Computing. Retrieved January 30, 2018 from http://www.computing.co.uk/
ctg/news/3001432/labour-party-to-call-for-regulation-of-technology-companies-
algorithms
Chambers, M., & Dinsmore, T. W. (2015). Advanced analytics methodologies: Driving busi-
ness value with analytics. Upper Saddle River, NJ: Pearson Education.
Chandler, N., Hostmann, B., Rayner, N., & Herschel, G. (2011). Gartner’s business analytics
framework. Stamford, CT: Gartner.
Chen, H., Chiang, R. H. L., & Storey, V. C. (2012). Business intelligence and analytics:
From big data to big impact. MIS Quarterly, 36(4), 1165–1188.
Chenoweth, M., Moore, N., Cox, A., Mele, J., & Sollinger, J. (2012). Best practices in sup-
plier relationship management and their early implementation in the Air Force materiel
command. Santa Monica, CA: The RAND Corporation.
Cohen, G., & Kotorov, R. (2016). Organizational intelligence: How smart companies use infor-
mation to become more competitive and profitable. New York, NY: Information Builders.
Davenport, T. H. (2013). Analytics 3.0. Harvard Business Review, 91(12), 64–72.
Davenport, T. H., & Dyché, J. (2013). Big data in big companies: International Institute
for Analytics. Retrieved January 30, 2018 from http://www.sas.com/resources/asset/
Big-Data-in-Big-Companies.pdf
28 ◾ Analytics and Knowledge Management

Davenport, T. H., & Harris, J. G. (2007). Competing on analytics: The new science of winning.
Boston, MA: Harvard Business School Review Press.
Davenport, T. H., & Short, J. E. (1990). The new industrial engineering: Information tech-
nology and business process redesign. Sloan Management Review 31(4), 11–27.
Donnelly, C. (2016). Machine learning helps Ocado’s customer services team wrap up
email overload. Computer Weekly. 29 November 2016, 10–12.
Edwards, J. S. (2009). Business processes and knowledge management. In M. Khosrow-
Pour (Ed.), Encyclopedia of information science and technology (2nd ed., Vol. I, pp.
471–476). Hershey, PA: IGI Global.
Edwards, J. S., & Moores, T. T. (1994). A conflict between the use of estimating and
planning tools in the management of information systems. European Journal of
Information Systems, 3(2), 139–147.
Edwards, J. S., & Rodriguez, E. (2016). Using knowledge management to give context
to analytics and big data and reduce strategic risk. Procedia Computer Science, 99,
36–49. doi:10.1016/j.procs.2016.09.099.
Fearnley, B. (2015). IDC MaturityScape: Big data and analytics in financial services.
Framingham, MA: IDC Research Retrieved January 30, 2018 from https://www.idc.
com/getdoc.jsp?containerId=US40619515.
Flinders, K. (2016). Case study: How Swedish bank prepared robot for customer services. In B.
Glick (Ed.), Artificial intelligence in the enterprise: Your guide to the latest thinking in AI and
machine learning (pp. 22–26): Computer Weekly. Retrieved January 30, 2018 from http://
www.computerweekly.com/ehandbook/Focus-Artificial-intelligence-in-the-enterprise.
Future of Life Institute. (2017). Asilomar AI principles. Retrieved February 11, 2017, from
https://futureoflife.org/ai-principles/
Goodwin, B. (2014). M&S turns to predictive analytics to keep shelves stocked over Christmas.
Computer Weekly. Retrieved January 30, 2018 from http://www.computerweekly.com/
news/2240236043/MS-turns-to-predictive-analytics-to-keep-shelves-stocked-over-Christmas
Haight, J., & Park, H. (2015). IoT analytics in practice. Boston, MA: Blue Hill Research.
Hall, P., Phan, W., & Whitson, K. (2016). The evolution of analytics: Opportunities and chal-
lenges for machine learning in business. Sebastopol, CA: O’Reilly Media.
Halper, F. (2015). Operationalizing and embedding analytics for action. Renton, WA: TDWI
Research.
Hammer, M. (1990). Re-engineering work: Don’t automate, obliterate. Harvard Business
Review, 68(4), 104–112.
Hawley, D. (2016). Implementing business analytics within the supply chain: Success and
fault factors. Electronic Journal of Information Systems Evaluation, 19(2), 112–120.
Henke, N., Bughin, J., Chui, M., Manyika, J., Saleh, T., Wiseman, B., & Sethupathy, G. (2016).
The age of analytics: Competing in a data-driven world: McKinsey Global Institute.
Retrieved January 30, 2018 from http://www.mckinsey.com/business-functions/
mckinsey-analytics/our-insights/the-age-of-analytics-competing-in-a-data-driven-world
Hill, K. (2012). How Target figured out a teen girl was pregnant before her father did.
Retrieved January 30, 2018 from https://www.forbes.com/sites/kashmirhill/2012/02/16/
how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/#5f0167436668
Holsapple, C., Lee-Post, A., & Pakath, R. (2014). A unified foundation for business analyt-
ics. Decision Support Systems, 64, 130–141.
Hong Kong Efficiency Unit. (2017). Efficiency unit. Retrieved March 12, 2017, from http://
www.eu.gov.hk/en/index.html
 Knowledge Management for Action-Oriented Analytics ◾ 29

Kirby, M. W., & Capey, R. (1998). The origins and diffusion of operational research in
the UK. Journal of the Operational Research Society, 49(4), 307–326. doi:10.1057/
palgrave.jors.2600558.
Kiron, D., Ferguson, R. B., & Prentice, P. K. (2013). From value to vision: Reimagining the
possible with data analytics. MIT Sloan Management Review, 54(3), 1–19.
Lopez-Rabson, T., & McCloy, U. (2013). Understanding student attrition in the six greater
Toronto area (GTA) colleges. Toronto, Canada: Higher Education Quality Council of
Ontario.
Luhn, H. P. (1958). A business intelligence system. IBM Journal of Research and
Development, 2(4), 314–319.
Marr, B. (2016). Big data at Tesco: Real time analytics at the UK grocery retail giant. Retrieved
January 30, 2018 from http://www.forbes.com/sites/bernardmarr/2016/11/17/
big-data-at-tesco-real-time-analytics-at-the-uk-grocery-retail-giant/#1c4d5cd4519a
Marr, B. (2017). IoT and big data at Caterpillar: How predictive maintenance saves
millions of dollars. Retrieved January 30, 2018 from https://www.forbes.com/sites/
bernardmarr/2017/02/07/iot-and-big-data-at-caterpillar-how-predictive-maintenance-
saves-millions-of-dollars/#7a18a9867240
McKenna, B. (2016a). Case study: Panama papers revealed by graph database visualisation
software. In B. McKenna (Ed.), Business intelligence in the world of big data (pp. 22–25):
Computer Weekly. Retrieved January 30, 2018 from http://www.computerweekly.
com/ehandbook/IT-Project-Business-intelligence-in-the-world-of-big-data.
McKenna, B. (2016b). Case study: The Christie speeds up SPC charts to improve clinical pro-
cesses. In B. McKenna (Ed.), Business intelligence in the world of big data (pp. 14–17):
Computer Weekly. Retrieved January 30, 2018 from http://www.computerweekly.
com/ehandbook/IT-Project-Business-intelligence-in-the-world-of-big-data.
Mezei, J., Brunelli, M., & Carlsson, C. (2016). A fuzzy approach to using expert knowl-
edge for tuning paper machines. Journal of the Operational Research Society, Advance
Online Publication. doi:10.1057/s41274-016-0105-3
MHR Analytics. (2017). Plotting the data journey in the boardroom: The state of data
analytics 2017. Nottingham, UK: MHR Analytics Retrieved January 30, 2018 from
http://www.mhr.co.uk/analytics
Moustakerski, P. (2015). Big data evolution: Forging new corporate capabilities for the long
term. London, UK: Economist Intelligence Unit.
Nexidia. (2014). Nexidia and Blue Cross and Blue Shield of North Carolina–voice of the
customer (VoC) analytics to increase clarity and ease of use for customers. Mountain
View, CA: Frost & Sullivan. Retrieved January 30, 2018 from http://www.nexidia.
com/media/2222/fs_casestudy-bcbsnc_final.pdf
Paulk, M. C., Curtis, B., Chrissis, M. B., & Weber, C. V. (1993). Capability maturity
model, version 1.1. IEEE Software, 10(4), 18–27.
Pechter, R. (2009). What’s PMML and what’s new in PMML 4.0? ACM SIGKDD
Explorations Newsletter, 11(1), 19–25.
Piskorski, M. J., Halaburda, H., & Smith, T. (2008). eHarmony. Cambridge, MA: Harvard
Business School. Retrieved January 30, 2018 from http://www.hbs.edu/faculty/Pages/
item.aspx?num=36554
Pittman, D. (2016). Big data, analytics and the retail industry: Luxottica. Retrieved
March 11, 2017, from http://www.ibmbigdatahub.com/presentation/big-data-
analytics-and-retail-industry-luxottica
30 ◾ Analytics and Knowledge Management

Polanyi, M. (1966). The tacit dimension. Garden City, NY: Doubleday.

Robinson, A., Levis, J., & Bennett, G. (2010). INFORMS news: INFORMS to officially join
analytics movement. OR/MS Today, 37(5). Retrieved January 30, 2018 from https://
www.informs.org/ORMS-Today/Public-Articles/October-Volume-37-Number-5/
INFORMS-News-INFORMS-to-Officially-Join-Analytics-Movement
Rodriguez, E. (Ed.). (2017). The analytics process: Strategic and tactical steps. Boca Raton,
FL: CRC Press.
SAS Institute. (2016a). Analytics for the future: The new “data generation”: Future
Foundation and SAS Institute. 1483298UK0916. Retrieved January 30, 2018 from
http://docs.media.bitpipe.com/io_13x/io_135639/item_1493324/future-of-analytics-
report-future-foundation%281%29.pdf
SAS Institute. (2016b). How any size organization can supersize results with data visualiza-
tion. Cary, NC: SAS Institute. Retrieved January 30, 2018 from https://www.sas.
com/en_us/whitepapers/any-size-business-can-supersize-results-with-data-visualization-
108199.html
Sims, M. (2016). Big data and analytics: Once a luxury, now a necessity. Houston, TX: American
Productivity and Quality Center. Retrieved January 30, 2018 from https://www.apqc.
org/knowledge-base/documents/big-data-and-analytics-once-luxury-now-necessity
Sopra Steria (2016). How to become an analytics powered enterprise: Key insights from the
early adopters: Sponsored by Sopra Steria. Retrieved January 30, 2018 from https://
www.soprasteria.com/docs/librariesprovider41/White-Papers/sopra-steria-computing-
data-analytics-wp-nov2016.pdf?status=Temp&sfvrsn=0.8211430164259691
Spender, J. C. (2007). Thinking, talking, and doing: How the knowledge-based view
might clarify technology’s relationship with organizations. International Journal of
Technology Management, 38(1/2), 178–196.
Stanford, D. (2013). Coke engineers its orange juice with an algorithm. Bloomberg News.
Retrieved April 24, 2017 from https://www.bloomberg.com/news/articles/2013-01-31/
coke-engineers-its-orange-juice-with-an-algorithm
TIBCO Software. (2015). Marks and Spencer empowers business analysts with Spotfire. Palo
Alto, CA: TIBCO Software. Retrieved January 30, 2018 from http://spotfire.tibco.
com/assets/blt3773b773ca2292eb/ss-marks-and-spencer.pdf
Virgo, P. (2017). Lost in the global cesspit of big data. Computer Weekly. Retrieved January
30, 2018 from http://www.computerweekly.com/blog/When-IT-Meets-Politics/
Lost-in-the-global-cesspit-of-Big-Data.
Watson, H., & Wixom, B. (2007). The current state of business intelligence. IEEE
Computer, 40(9), 96–99.
White, E., & Jordan, T. (2016). National Audit Office framework to review models.
(pp. 11018–002). London, UK.
Witchalls, C. (2014). Gut & gigabytes: Capitalising on the art & science in decision
making: PwC. Retrieved April 24, 2017 from http://www.pwc.com/gx/en/issues/
data-and-analytics/big-decisions-survey/assets/big-decisions2014.pdf
Chapter 2

Data Analytics Process:

An Application
Case on Predicting
Student Attrition
Dursun Delen

Contents
Introduction to Data Analytics Processes .............................................................32
Knowledge Discovery in Databases Process .....................................................32
Cross-Industry Standard Process for Data Mining ......................................... 34
Step 1—Business Understanding................................................................35
Step 2—Data Understanding .....................................................................35
Step 3—Data Preparation ..........................................................................36
Step 4—Model Building ............................................................................37
Step 5—Testing and Evaluation .................................................................38
Step 6 —Deployment ................................................................................39
Sample, Explore, Modify, Model, Assess Process .............................................39
Step 1—Sample ........................................................................................ 40
Step 2—Explore .........................................................................................41
Step 3—Modify .........................................................................................41
Step 4—Model...........................................................................................41
Step 5—Assess .......................................................................................... 42

31
32 ◾ Analytics and Knowledge Management

Six Sigma for Data Analytics .......................................................................... 42

Step 1—Define ......................................................................................... 44
Step 2—Measure ....................................................................................... 44
Step 3—Analyze .........................................................................................45
Step 4—Improve ........................................................................................45
Step 5—Control ........................................................................................45
Which Process Is the Best? ..............................................................................45
Application Case: Predicting Student Attrition with Data Analytics ................... 46
Introduction and Motivation ..........................................................................47
Analytics Methodology ...................................................................................49
Data Description........................................................................................50
Predictive Analytics Models ........................................................................54
Sensitivity Analysis .....................................................................................56
Results ............................................................................................................57
Discussion and Conclusions ...........................................................................59
References ...........................................................................................................63

Introduction to Data Analytics Processes

Despite the popularity of data analytics in this age of Big Data, there is still not
a proven, standardized, and universally accepted process to conduct data analytics
projects. As has been the case in any previous computational paradigms, discovering
or creating knowledge from large and diverse data repositories (i.e., Big Data) started
as trial-and-error type experimental initiatives. Many have looked at the problem
from a different, relatively narrow perspective, trying to characterize what does and
doesn’t work for their specific problem and purpose. For quite some time, data ana-
lytics projects were carried out as experimental or trial-and-error endeavors. However,
to systematically conduct data analytics, a standardized process needed to be devel-
oped and regularly followed and practiced. Based on the best practices, data analytics
researchers and practitioners have proposed several processes (i.e., workflows, in the
form of simple step-by-step methodologies) to maximize the possibility of success
in conducting data analytics projects. These efforts—mostly modeled after the data
mining and knowledge discovery methodologies—resulted in a few standardized
processes, of which the most popular ones are described in the following subsections.

Knowledge Discovery in Databases Process

One of the earliest and perhaps the first data mining (i.e., predictive analytics) pro-
cess is proposed by Fayyad et al. (1996) under the name of knowledge discovery in
databases (KDD) methodology. In their seminal work, Fayyad and his colleagues
proposed KDD as a methodology within which data mining is believed to be a
single step or task where the novel patterns are identified and extracted from data.
They proposed KDD as a comprehensive end-to-end process that encompasses
 Data Analytics Process ◾ 33

Internalization

Data mining
Knowledge
Data 5
1 2 3 4
transformation
Extracted
patterns
Data
cleaning
Transformed
data
Data
selection Preprocessed
data
Target
data

Feedback
Raw data

Figure 2.1 Knowledge discovery in databases (KDD) process.

many individual steps or tasks to convert data into knowledge (i.e., actionable
insight). A pictorial representation of the KDD process is given in Figure 2.1.
In Figure 2.1, the processing steps are shown as directed arrows with callout
labels, and the result of each step is shown as a graphical image representing the arti-
fact. As shown therein, the input to the KDD process is a collection of data coming
from organizational databases and/or other external mostly structured data sources.
These data sources are often combined in a centralized data repository called a data
warehouse. A data warehouse enables the KDD process to be implemented effec-
tively and efficiently because it provides a single source for data to be mined. Once
the data are consolidated in a unified data warehouse, the problem-specific data
is extracted and prepared for further processing. As the data are usually in a raw,
incomplete, and dirty state, a through preprocessing need to be conducted before
the modeling can take place. Once the data are preprocessed and transformed into
a form for modeling, a variety of modeling techniques are applied to the data to
convert it into patterns, correlations, and predictive models. Once the discovered
patterns are validated, they need to be interpreted and internalized so that they can
be converted into actionable information (i.e., knowledge). One important part of
this process is the feedback loop that allows the process flow to redirect backward,
from any step to any other previous steps, for rework and readjustments.
34 ◾ Analytics and Knowledge Management

Cross-Industry Standard Process for Data Mining

Another standardized analytics process, arguably the most popular one, is called
Cross-Industry Standard Process for Data Mining (CRISP-DM), which was pro-
posed in the mid to late 1990s by a European consortium of companies. It was
intended to serve as a nonproprietary standard methodology for data mining proj-
ects (Chapman et al., 2000). Figure 2.2 illustrates this six step standardized pro-
cess, which starts with a good understanding of the business problem and the need
or objective for the data mining project (i.e., the application domain) and ends with
the deployment of the solution that satisfied the specific business need that started
the data mining project in the first place (Delen, 2015). Even though these steps are
shown sequentially in the graphical representation, there usually is a great deal of
backtracking and feedback. Because the data mining is driven by experience and
experimentation, depending on the problem situation and the skills, knowledge,
and experience of the analyst, the whole process can be very iterative (i.e., one that
would require going back and forth through the steps quite a few times) and time
demanding. Because a latter step is built on the outcome of the immediate prede-
cessor, one should pay extra attention to the earlier steps in order not to put the
whole study on an incorrect path from the onset. What follows is a relatively brief
description of each step in CRISP-DM methodology.

1 2
Business Data
understanding understanding

3
Data
preparation
6
4
Deployment
Model
Data
building

5
Testing and
evaluation

Figure 2.2 CRISP-DM data mining process.

 Data Analytics Process ◾ 35

Step 1—Business Understanding

The key to success in any analytics project is to know what the study is meant for
(i.e., the purpose or the objective). Answering this question begins with a thor-
ough understanding of the managerial need for new knowledge and an explicit
specification of the business objective regarding the study to be conducted.
Specific goals such as “What are the common characteristics of the customers we
have lost to our competitors recently?” or “What are typical profiles of our cus-
tomers, and how much value does each of them provide to us?” are needed. Then
a project plan for finding such knowledge is developed that specifies the people
responsible for collecting the data, analyzing the data, and reporting the findings.
At this early stage, a budget to support the study should also be established, at
least at a high level with rough numbers. For example, in a customer segmenta-
tion model, developed for a retail catalog business, the identification of a business
purpose meant identifying the type of customer that would be expected to yield a
profitable return. The same or similar analysis would also be useful for credit card
distributors. For business purposes, grocery stores often try to identify which
items tend to be purchased together so it can be used for affinity positioning
within the store, or to intelligently guide promotional campaigns. Data mining
has many useful business applications applied to many business problems and
opportunities. Intimately knowing the business purpose is very critical to achiev-
ing success.

Step 2—Data Understanding

There is a need to make a perfect match between the business problem and the
data being used to address it. That is, an analytics study is specific to addressing a
well-defined business task, and different business tasks require different sets of data.
Following the business understanding, the next activity in the data mining process
is to identify the relevant data from many available data sources. There are several
key points that must be considered in the data identification and selection process.
First and foremost, the analyst should be clear and concise about the description of
the data mining task so that the most relevant data can be identified. For example,
a retail data mining project may seek to identify spending behaviors of female shop-
pers who purchase seasonal clothes based on their demographics, credit card trans-
actions, and socioeconomic attributes. Furthermore, the analyst should build an
intimate understanding of the data sources (e.g., where the relevant data are stored
and in what form; what the process of collecting the data is—automated versus
manual; who the collectors of the data are and how often the data are updated)
and the variables (e.g., what the most relevant variables are, if there are any synony-
mous or homonymous variables, if the variables are independent of each other—
do they stand as a complete information source without overlapping or conflicting
information).
36 ◾ Analytics and Knowledge Management

To better understand the data, the data scientist often uses a variety of statistical
and graphical tools and techniques, such as simple statistical descriptors or summa-
ries of each variable (e.g., for numeric variables the average, minimum, maximum,
median, and standard deviation are among the calculated measures, whereas for
categorical variables, the mode and frequency tables are calculated), correlation
analysis, scatterplots, histograms, box plots, and cross tabulation. A through pro-
cess of identification and selection of data sources and the most relevant variables
can make it more straightforward for downstream algorithms to quickly and accu-
rately discover useful knowledge patterns.
Data sources for data selection can vary. Normally, data sources for business
applications include demographic data (such as income, education, number of
people in a household, and age), sociographic data (such as hobby, club member-
ship, and entertainment), and transactional data (such as sales record, credit card
spending, and issued checks), among others. Regardless of the sources, data can be
categorized as quantitative and qualitative. Quantitative data are measured using
numeric values. It can be discrete (such as integers) or continuous (such as real
numbers). Qualitative data, also known as categorical data, contain both nominal
and ordinal data. Nominal data have finite nonordered values (e.g., gender data,
which has two values: male and female). Ordinal data have finite ordered values.
For example, customer credit ratings are considered ordinal data because the ratings
can be excellent, fair, and bad. Quantitative data can be readily represented by some
sort of probability distribution. A probability distribution describes how the data are
dispersed and shaped. For instance, normally distributed data are symmetric and is
commonly referred to as being a bell-shaped curve. Qualitative data may be coded
to numbers and then described by frequency distributions. Once the relevant data
are selected according to the analytics project objectives, another critical task—data
preparation and preprocessing—would be conducted.

Step 3—Data Preparation

The purpose of data preparation (or more commonly called data preprocessing) is
to take the data identified in the previous step and prepare it for analytics methods
and algorithms. Compared to the other steps in CRISP-DM, data preprocessing
consumes the most amount of time and effort—most believe that this step accounts
for more than 80% of the total time spent on an analytics project. The reason for
such an enormous effort spent on this step is the fact that real-world data are gener-
ally incomplete (lacking attribute values, lacking certain attributes of interest, or
containing only aggregate data), noisy (containing errors or outliers), and inconsis-
tent (containing discrepancies in codes or names).
Some parts of the data may have different formats because they are taken from
different data sources. If the selected data are from flat files, voice message, images,
 Data Analytics Process ◾ 37

and web pages, they need to be converted to a consistent and unified format. In gen-
eral, data cleaning means to filter, aggregate, and fill in missing values (also known
as imputation). By filtering the data, the selected variables are examined for outliers
and redundancies. Outliers differ greatly from the majority of data, or data that are
clearly out of range of the selected data groups. For example, if the age of a customer
included in the data is 190, it should be a data entry error and should be identified
and fixed (perhaps, taken out from the data mining project that examines the vari-
ous aspects of the customers and age is perceived to be a critical component of the
customer characteristics). Outliers may be caused by many reasons, such as human
errors or technical errors, or may naturally occur in a dataset due to extreme events.
Suppose the age of a credit card holder is recorded as “12.” This is likely a data entry
error, most likely by a human. However, there might actually be an independently
wealthy pre-teen with important purchasing habits. Arbitrarily deleting this outlier
could dismiss valuable information.
Redundant data are the same information recorded in several different ways.
Daily sales of a particular product are redundant to seasonal sales of the same
product, because we can derive the sales from either daily data or seasonal data. By
aggregating data, data dimensions are reduced to obtain aggregated information.
Note that although an aggregated dataset has a small volume, the information will
remain. If a marketing promotion for furniture sales is considered in the next three
or four years, then the available daily sales data can be aggregated as annual sales
data. The size of sales data is dramatically reduced. By smoothing data, missing val-
ues of the selected data are found and new or reasonable values then added. These
added values could be the average number of the variable (mean) or the mode.
A missing value often causes no solution when a data-mining algorithm is applied
to discover the knowledge patterns.

Step 4—Model Building

In this step, various modeling methods and algorithms are designed and applied
to an already prepared dataset to address the specific business question, need, or
purpose. The modeling step also encompasses the assessment and comparative
analysis of the various types of models that can address the same type of data min-
ing tasks (e.g., clustering, classification, etc.). Because there is not a universally
known best method or algorithm for a specific data mining task, one should use
a variety of viable model types along with a well-defined experimentation and
assessment strategy to identify the “best” method for a given data mining problem.
Even for a single method or algorithm, a number of parameters must be calibrated
to obtain optimal results. Some methods may have specific requirements on the
way that the data are to be formatted; thus stepping back to the data preparation
step is often necessary.
38 ◾ Analytics and Knowledge Management

Depending on the business need, the analytics task can be of a prediction (either
classification or regression), an association, or a clustering or segmentation type.
Each of these tasks can use a variety of analytics and data mining methods and
algorithms. For instance, classification type data mining tasks can be accomplished
by developed neural networks, decision trees, support vector machines (SVMs), or
logistic regression, among others.
The standard procedure for modeling in data mining is to take a large prepro-
cessed dataset and divide it into two or more subsets for training and validation
or testing. Then, use a portion of the data (the training set) for development of
the models (no matter what modeling technique or algorithms is used), and use
the other portion of the data (the test set) for testing the model that is just built.
The principle is that if you build a model on a particular set of data, it will of course
test quite well on the data that is was built on. By dividing the data and using part
of it for model development and a separate part of it for testing creates objective
results for the accuracy and reliability of the model. The idea of splitting the data
into components is often carried to additional levels with multiple splits in the
practice of data mining. Further details about data splitting and other evaluation
methods can be found in Delen (2015).

Step 5—Testing and Evaluation

In this step, the developed models are assessed and evaluated for their accuracy and
generality. This step assesses the degree to which the selected model (or models)
meets the business objectives and, if so, to what extent (i.e., do more models need to
be developed and assessed). Another option is to test the developed models in a real-
world scenario if time and budget constraints permit. Even though the outcome of
the developed models is expected to relate to the original business objectives, other
findings that are not necessarily related to the original business objectives, but that
might also unveil additional information or hints for future directions, often are
discovered.
The testing and evaluation step is a critical and challenging task. No value
is attained by an analytics project unless the expected business outcomes are
achieved through the newly discovered knowledge patterns. Determining the
business value from discovered knowledge patterns is somewhat similar to play-
ing with puzzles. The extracted knowledge patterns are pieces of the puzzle that
need to be put together in the context of the specific business purpose. The suc-
cess of this identification depends on the interaction among data scientists, busi-
ness analysts, and decision makers (such as business managers). Because data
scientists and analysts may not have the holistic understanding of the business
objectives and what they mean to the business operations, and the business ana-
lysts and decision makers may not have the technical knowledge to interpret the
results often presented in seemingly complex mathematical or numerical format,
interaction among them is crucial. To properly interpret knowledge patterns,
 Data Analytics Process ◾ 39

it is often necessary to use a variety of tabulation and visualization techniques

(e.g., pivot tables, cross tabulation of findings, pie charts, histograms, box plots,
scatter plots, etc.).

Step 6—Deployment
Development and assessment of the models is not the end of the analytics project.
Even if the purpose of the model is to have a simple exploration of the data, the
knowledge gained from such exploration will need to be organized and presented
in a way that the end user can understand and benefit from it. Depending on the
requirements, the deployment phase can be as simple as generating a report or
as complex as implementing a repeatable computer-based decision support system
across the enterprise (Delen, Sharda, and Kumar, 2007). In many cases, it is the
customer, not the data analyst, who carries out the deployment steps. However,
even if the analyst will not carry out the deployment effort, it is important for the
customer to understand up front what actions need to be carried out to actually
make use of the created models.
The deployment step may also include maintenance activities for the deployed
models. Because everything about the business is constantly changing, the data that
reflect the business activities also is changing. Over time, the models (and the pat-
terns embedded within them) built on the old data becomes obsolete, irrelevant, or
misleading. Therefore, monitoring and maintenance of the models are important,
if the analytics results are to become a part of the day-to-day business decision-
making environment. A careful preparation of a maintenance strategy helps to
avoid unnecessarily long periods of incorrect usage of analytics results. To monitor
the deployment of the analytics results, the project needs a detailed plan on the
monitoring process, which may not be a trivial task for complex analytics models.
The CRISP-DM process is the most complete and most popular data mining
methodology being practiced in industry as well as in academia. As opposed to
using it as is, practitioners add their own insight to make it specific to their organi-
zation’s style of practice.

Sample, Explore, Modify, Model, Assess Process

In order to be applied successfully, the data analytics projects must be viewed as a
process rather than a set of tools or techniques. In addition to the CRISP-DM pro-
cess, there is yet another well-known methodology developed by the SAS Institute,
called sample, explore, modify, model, assess (SEMMA) process. Beginning with a
statistically representative sample of your data, SEMMA intends to make it easy to
apply exploratory statistical and visualization techniques, select and transform the
most signify cant predictive variables, model the variables to predict outcomes, and
finally confirm a model’s accuracy. A pictorial representation of SEMMA is given
in Figure 2.3.
40 ◾ Analytics and Knowledge Management

Sample
(Generate a representative
SEMMA sample of the data)

Assess Explore
(Evaluate the accuracy and (Visually explore and
usefulness of the models) describe the data)

Feedback

Model Modify
(Use a variety of statistical and (Select variables, transform
machine learning models) variable representations)

Figure 2.3 A graphical depiction of the SEMMA process.

By assessing the outcome of each stage in the SEMMA process, one can
determine how to model new questions raised by the previous results, and thus
proceed back to the exploration phase for additional refinement of the data.
That is, as is the case in CRISP-DM, SEMMA also driven by a highly iterative
experimentation cycle. Here are short descriptions for the five steps that consti-
tute SEMMA.

Step 1—Sample
This is where a portion of a large dataset (big enough to contain the significant
information yet small enough to manipulate quickly) is extracted. For optimal
cost and computational performance, some (including the SAS Institute) advocate
a sampling strategy, which applies a reliable, statistically representative sample
of the full detail data. In the case of very large datasets, mining a representa-
tive sample instead of the whole volume may drastically reduce the processing
time required to get crucial business information. If general patterns appear in
the data as a whole, these will be traceable in a representative sample. If a niche
(rare pattern) is so tiny that it is not represented in a sample and yet so important
that it influences the big picture, it should be discovered using exploratory data
 Data Analytics Process ◾ 41

description methods. It is also advised to create partitioned datasets for better

accuracy assessment.

◾ Training, for model fitting

◾ Validation, for assessment and to prevent over fitting
◾ Test, for obtaining an honest assessment of how well a model generalizes

A more detailed discussion and relevant techniques for assessment and validation of
data mining models can be found in Sharda et al. (2017).

Step 2—Explore
This is where a user searches for unanticipated trends and anomalies to gain a
better understanding of the dataset. After sampling your data, the next step is to
explore it visually or numerically for inherent trends or groupings. Exploration
helps refine and redirect the discovery process. If visual exploration does not reveal
clear trends, one can explore the data through statistical techniques including
factor analysis, correspondence analysis, and clustering. For example, in data min-
ing for a direct mail campaign, clustering might reveal groups of customers with
distinct ordering patterns. Limiting the discovery process to each of these distinct
groups individually may increase the likelihood of exploring richer patterns that
may not be strong enough to be detected if the whole dataset is to be processed
together.

Step 3—Modify
This is where the user creates, selects, and transforms the variables upon which
to focus the model construction process. Based on the discoveries in the explora-
tion phase, one may need to manipulate data to include information such as the
grouping of customers and significant subgroups, or to introduce new variables.
It may also be necessary to look for outliers and reduce the number of variables,
to narrow them down to the most significant ones. One may also need to modify
data when the “mined” data changes. Because data mining is a dynamic, iterative
process, you can update data mining methods or models when new information
is available.

Step 4—Model
This is where the user searches for a variable combination that reliably predicts a
desired outcome. Once you prepare your data, you are ready to construct models
that explain patterns in the data. Modeling techniques in data mining include
artificial neural networks (ANN), decision trees, rough set analysis, SVMs, logistic
models, and other statistical models—such as time series analysis, memory-based
42 ◾ Analytics and Knowledge Management

reasoning, and principal component analysis. Each type of model has particular
strengths, and is appropriate within specific data mining situations depending on
the data. For example, ANN are very good at fitting highly complex nonlinear
relationships while rough sets analysis is known to produce reliable results with
uncertain and imprecise problem situations.

Step 5—Assess
This is where the user evaluates the usefulness and the reliability of findings from
the data mining process. In this final step of the data mining process, the user
assesses the model to estimate how well it performs. A common means of assessing
a model is to apply it to a portion of a dataset put aside (and not used during the
model building) during the sampling stage. If the model is valid, it should work
for this reserved sample as well as for the sample used to construct the model.
Similarly, you can test the model against known data. For example, if you know
which customers in a file had high retention rates and your model predicts reten-
tion, you can check to see whether the model selects these customers accurately. In
addition, practical applications of the model, such as partial mailings in a direct
mail campaign, help prove its validity.
The SEMMA process is quite compatible with the CRISP-DM process. Both
aim to streamline the knowledge discovery process. Both were created as broad
frameworks, which need to be adapted to specific circumstances. In both, once
models are obtained and tested, they can then be deployed to gain value with
respect to business or research application. Even though they have the same goal
and are similar, the SEMMA and CRISP-DM processes have a few differences.
Table 2.1 presents these differences.

Six Sigma for Data Analytics

Six Sigma is a popular business management philosophy that focuses on reducing
the deviations (i.e., sigma) from the perfection by employing rigorous and system-
atic execution of proven quality control principles and techniques. This vastly pop-
ular management philosophy was first introduced by Motorola in the 1980s in the
context of manufacturing management and has since then been adopted by many
companies and organizations in a wide variety of business contexts beyond manu-
facturing. Ideally, it promotes zero defects and zero tolerance, which can roughly
be translated in the business context as error free, perfect business execution. Six
Sigma methodology manifested itself in the business world with the define, mea-
sure, analyze, improve, and control (DMAIC) process. Because of its success in
many business problems and settings, DMAIC methodology is also applied to
analytics projects. Figure 2.4 shows a DMAIC methodology as a simplified flow
diagram.
Table 2.1 Comparison of CRISP-DM and SEMMA
Task CRISP-DM SEMMA Comments

Project initiation Business understanding N/A In this phase, CRISP-DM includes activities like project initiation,
problem definition, and goal setting. SEMMA does not have a
step for this phase.

Data access Data understanding Sample In this phase, both CRISP-DM and SEMMA have the steps to
Explore access, sample, and explore the data.

Data transformation Data preparation Modify In this phase, both CRISP-DM and SEMMA process the data to
make it amenable to machine processing.

Model building Modeling Model In this phase, both CRISP-DM and SEMMA suggest building and
testing various models.

Project evaluation Evaluation Assess In this phase, both CRISP-DM and SEMMA suggest assessing the
findings against the project goals.

Project finalization Deployment N/A In this phase, CRISP-DM prescribes deployment of the results
while SEMMA does not explicitly state it.
Data Analytics Process
◾
43
44 ◾ Analytics and Knowledge Management

Step 1
Define
(Understand the problem and
define the project goals)

Step 2

Measure
(Measure the suitability and
quality of data)

Step 3

Analyze
(Experiment with different
models to identify the best fit)

Step 4

Improve
(Assess the knowledge
against the project goals)

Step 5
Control
(Deploy the solutions and
Feedback control its usability)

Figure 2.4 Six Sigma DMAIC methodology.

Step 1—Define
This is the first step in DMAIC where several tasks are to be accomplished to get
the project set up and started. These tasks include (1) a thorough understanding of
the business needs; (2) identifying the most pressing problem, (3) defining the goals
and objectives, (4) identifying and defining the data and other resources needed to
investigate the business problem, and (5) developing a detailed project plan. As you
may have noticed, there is a significant overlap between this step and the “Business
Understanding” which was the first step in the CRISP-DM process.

Step 2—Measure
In this step, the mapping between organizational data repositories and the business
problem is assessed. Since data mining requires problem-relevant, clean, and usable
data, identification and creation of such a resource is of critical importance to the
success of the project. In this step, the identified data sources are to be consolidated
and transformed into a format that is amenable to machine processing.
 Data Analytics Process ◾ 45

Step 3—Analyze
Now that the data are prepared for processing, in this step, a series of data mining
techniques is used to develop models. Since there is not a single best technique for
a specific data mining task (because there are many and most of them are machine
learning techniques with many parameters to optimize), several probable tech-
niques need to be applied and experimented with to identify and develop the most
appropriate model.

Step 4—Improve
Once the analysis results are obtained, in this step, the improvement possibilities
are investigated. Improvements can be at the technique level or they can be at the
business problem level. For instance, if the model results are not satisfactory, other
more sophisticated techniques (e.g., ensemble systems) can be used to boost the
performance of the models. Also, if the modeling results are not clearly addressing
the business problem, via a feedback loop to previous steps, the very structure of the
analysis can be re-examined and improved, or the business problem can be further
investigated and restated.

Step 5—Control
In this step a final examination of the project outcomes is assessed and if found
satisfactory, the models and result are disseminated to decision makers and/or inte-
grated into the existing business intelligence systems for automation.
The Six Sigma-based DMAIC methodology has a lot of resemblance to the
CRISP-DM process. We do not have any evidence that suggests one is inspired
from the other. That said, since what these two processes portray are rather logical
and straightforward steps in any business system analysis effort, they may not have
inspired each other. The users of DMAIC are rare compared to CRISP-DM and
SEMMA.

Which Process Is the Best?

Even though some are more elaborate than others, there is not a sure way to com-
pare these data analytics processes. They all have pros and cons in their own respect.
Some are more problem focused, while others are more analysis driven. Businesses
that does data mining adopts one of these methodologies, and often slightly modi-
fies it to make it their own. To shed some more light to the question of “Which
one is the best?” KDNuggets.com, a well-known and well-respected web portal for
data mining, conducted a survey where it asked the very same question. Figure 2.5
shows the polling results of that survey (Piatetsky, 2014).
As the survey results indicate, CRISP-DM dominates the others as the most
popular data mining process. Compared to the others, CRISP-DM is the most
46 ◾ Analytics and Knowledge Management

4.7%
A domain-specific methodology
2.0%

5.3%
My organizations'
3.5%

7.3%
KDD process
7.5%

Other, not domain-specific 4.0%

8.0%

13.0%
SEMMA
8.5%

19.0%
My own
27.5%

42.0%
CRISP-DM
43.0%
0.0% 10.0% 20.0% 30.0% 40.0% 50.0%

2007 Poll 2014 Poll

Figure 2.5 Preference poll for standard analytics processes. (From Piatetsky, G.,
CRISP-DM, still the top methodology for analytics, data mining, or data science
projects, KDnuggets. Retrieved from http://www.kdnuggets.com/2014/10/crisp-
dm-top-methodology-analytics-data-mining-data-science-projects.html, 2014.)

complete and most matured methodology for data analytics projects. Many of the
ones that fall under “My Own” are also known to be small deviations (specializations)
of CRISP-DM.

Application Case: Predicting Student Attrition

with Data Analytics
Predicting attrition has always been an intriguing and challenging problem for
data scientists and business managers. Attrition (a tool to better predict and man-
age retention) is an important subject in many application domains, and based
on the domain, it may be called differently. For instance, in marketing, it is often
called “customer churn”, referring to the timely identification of at-risk customers
(i.e., one about to leave your products or services). The marketing saying that goes
something like “attracting a new customer cost ten times more than retaining the
one you have” is a testament to the importance of predicting and properly manag-
ing churn. Attrition is also an important concept in employee retention, where
accurate prediction and management of at-risk employees would save time and
money for the organization in maintaining a productive and capable workforce.
 Data Analytics Process ◾ 47

In this application case, the focus will be on predicting student attrition for better
management of student retention in higher education institutions.
Student retention is a critical part of many enrollment management systems.
Affecting university rankings, school reputation, and financial well-being, student
retention has become one of the most important priorities for decision makers in
higher education institutions. Improving student retention starts with a thorough
understanding of the reasons behind the attrition. Such an understanding is the
basis for accurately predicting at-risk students and appropriately intervening to
retain them. In this study, using five years of institutional data along with several
data mining techniques (both individuals as well as ensembles), we developed ana-
lytical models to predict and to explain the reasons behind freshmen student attri-
tion. The comparative analysis results showed that the ensembles performed better
than the individual models, while a balanced dataset produced better prediction
results than an unbalanced dataset. The sensitivity analysis of the models revealed
that the educational and financial variables are among the most important predic-
tors of the phenomenon.

Introduction and Motivation

Student attrition has become one of the most challenging problems for decision
makers in academic institutions. In spite of all of the programs and services to
help retain students, according to the U.S. Department of Education, Center for
Educational Statistics (nces.ed.gov), only about half of those who enter higher edu-
cation actually earn a bachelor’s degree. Enrollment management and the retention
of students has become a top priority for administrators of colleges and universities
in the United States and other developed countries around the world. A high stu-
dent drop-out rate usually results in overall financial loss, lower graduation rates,
and inferior school reputation in the eyes of all stakeholders (Gansemer-Topf and
Schuh, 2006). The legislators and policymakers who oversee higher education and
allocate funds, the parents who pay for their children’s education to prepare them
for a better future, and the students who make college choices look for evidence
of institutional quality and reputation to guide their decision-making processes
(Delen, 2010).
The principal motivation for improving student retention is the economic and
social benefits of attaining a higher education degree (Thomas and Galambos,
2004), both for individuals and for the public. Generally speaking, the economic
and social attributes that motivate individuals to enter higher education are
(1) public economic benefits, such as increased tax revenues, greater productivity,
increased consumption, increased workforce flexibility, and decreased reliance on
government financial support; (2) individual economic benefits such as higher sala-
ries and benefits, employment, higher savings levels, improved working conditions,
and personal and professional mobility; (3) public social benefits, such as reduced
crime rates, increased charitable giving and community service, increased quality
48 ◾ Analytics and Knowledge Management

of civic life, social cohesion and the appreciation of diversity, and the improved
ability to adapt to and use technology; and (4) individual social benefits, such as
improved health and life expectancy, improved quality of life for offspring, better
consumer decision-making, increased personal status, and more hobbies and leisure
activities (Hermaniwicz, 2003)
Traditionally, student attrition at a university has been defined as the num-
ber of students who do not complete a degree in that institution. Studies have
shown that the vast majority of students withdraw during their first year of col-
lege than during the rest of their higher education (Deberard, Julka, and Deana,
2004; Hermaniwicz, 2003). Since most of the student dropouts occur at the end
of the first year (the freshman year), many of the student retention and attrition
studies (including this study) have focused on first-year dropouts (or the number
of students not returning for the second year). This definition of attrition does not
differentiate between the students who may have transferred to other universities
and obtained their degrees there. It only considers the students dropping out at the
end of the first year voluntarily and not by academic dismissal.
Research on student retention has traditionally been survey driven (e.g., sur-
veying a student cohort and following them for a specified period of time to deter-
mine whether they continue their education) (Caison, 2007). Using such a design,
researchers worked on developing and validating theoretical models including
the famous student integration model developed by Tinto (1993). Elaborating on
Tinto’s theory, others have also developed student attrition models using survey-
based research studies (Berger and Braxton, 1998; Berger and Milem, 1999). Even
though they have laid the foundation for the field, these survey-based research
studies have been criticized for their lack of generalized applicability to other insti-
tutions and the difficulty and costliness of administering such large-scale survey
instruments (Cabrera, Nora, and Castaneda, 1993). An alternative (and/or a com-
plementary) approach to the traditional survey-based retention research is an ana-
lytic approach where the data commonly found in institutional databases is used.
Educational institutions routinely collect a broad range of information about their
students, including demographics, educational background, social involvement,
socioeconomic status, and academic progress. A comparison between the data-
driven and survey-based retention research showed that they are comparable at
best, and to develop a parsimonious logistic regression model, data-driven research
was found to be superior to its survey-based counterpart (Caison, 2007). But in
reality, these two research techniques (one driven by surveys and theories and the
other driven by institutional data and analytic methods) complement and help each
other (Miller and Tyree, 2009). That is, the theoretical research may help identify
important predictor variables to be use in analytical studies while analytical studies
may reveal novel relationships among the variables that may lead to development of
new and betterment of the existing theories.
To improve student retention, one should try to understand the non-trivial rea-
sons behind the attrition. To be successful, one should also be able to accurately
 Data Analytics Process ◾ 49

identify those students that are at risk of dropping out. So far, the vast majority
of student attrition research has been devoted to understanding this complex, yet
crucial, social phenomenon. Even though these qualitative, behavioral, and survey-
based studies revealed invaluable insight by developing and testing a wide range of
theories, they do not provide the much-needed instrument to accurately predict
(and potentially improve) student attrition (Delen, 2011; Miller and Herreid, 2010;
Veenstra, 2009). In this project, we propose a quantitative research approach where
the historical institutional data from student databases is used to develop models
that are capable of predicting, as well as explaining, the institution-specific nature
of the attrition problem. Though the concept is relatively new to higher educa-
tion, for almost a decade now, similar problems in the field of marketing have
been studied using predictive data mining techniques under the name of “churn
analysis,” where the purpose is to identify among the current customers who are
most likely to leave the company so that some kind of intervention process can be
executed for the ones who are worthwhile to retain. Retaining existing customers
is crucial because the related research shows that acquiring a new customer costs
roughly ten times more than keeping the one that you already have (Lemmens and
Croux, 2006).

Analytics Methodology
In this research, we followed a popular data mining methodology called CRISP-DM
(Shearer, 2000), that, as explained in the previous section, is a six step process: (1)
understanding the domain and developing the goals for the study, (2) identifying,
accessing, and understanding the relevant data sources, (3) preprocessing, clean-
ing, and transforming the relevant data, (4) developing models using comparable
analytical techniques, (5) evaluating and assessing the validity and the utility of the
models against each other and against the goals of the study, and (6) deploying the
models for use in decision-making processes. This popular methodology provides
a systematic and structured way of conducting data mining studies, and hence
increasing the likelihood of obtaining accurate and reliable results. The attention
paid to the earlier steps in CRISP-DM (i.e., understanding the domain of study,
understanding the data, and preparing the data) sets the stage for a successful data
mining study. Roughly 80% of the total project time is usually spent on these first
three steps.
The method evaluation step in CRISP-DM requires comparing the data mining
models for their predictive accuracy. Traditionally, in this comparison process the
complete dataset is split into two subsets, two-thirds for training and one-third for
testing. The models are trained on the training subset and then evaluated on the
testing subset. The prediction accuracy on the testing subset is used to report the
actual prediction accuracies of all evaluated models. Since the dataset is split into
two exclusive subsets randomly, there always is a possibility of those two datasets
not being “equal.” To minimize this bias associated with the random sampling
50 ◾ Analytics and Knowledge Management

of the training and testing data samples, we used an experimental design called
k-fold cross-validation. In k-fold cross-validation, also called rotation estimation,
the complete dataset is randomly split into k mutually exclusive subsets of approxi-
mately equal size. The classification model is trained and tested k times. Each time,
it is trained on all but one fold and tested on the remaining single fold. The cross-
validation estimate of the overall accuracy is calculated as simply the average of the
k individual accuracy measures as in the following equation:
k

∑ PM
1
CV = i
k i =1

where, CV stands for the cross-validation result for a method, k is the number of
folds used, and PM is the performance measure for each fold.
In this case study, to estimate the performance of the prediction models, a ten-
fold cross-validation approach was used. Empirical studies showed that 10 seems
to be an optimal number of folds (that optimizes the time it takes to complete the
test while minimizing the bias and variance associated with the validation process)
(Kohavi, 1995). In tenfold cross-validation the entire dataset is divided into 10
mutually exclusive subsets (or folds). Each fold is used once to test the performance
of the prediction model that is generated from the combined data of the remaining
ninefolds, leading to 10 independent performance estimates.
A pictorial depiction of this evaluation process is shown in Figure 2.6. With this
experimental design, if k is set to 10 (which is the case in this study and a common
practice in most predictive data mining applications), for each of the seven model
types (four individual and three ensembles), 10 different models are developed and
tested. Combined with the tenfold experimentation conducted on the original (i.e.,
unbalanced) datasets using the four individual model types, the total number of
models developed and tested for this study was 110.

Data Description
The data for this study came from a single institution (a comprehensive public uni-
versity located in the Midwest region of the United States) with an average enroll-
ment of 23,000 students, of which roughly 80% are residents of the same state and
roughly 19% of the students are listed under some minority classification. There is
no significant difference between the two genders in the enrollment numbers. The
average freshman student retention rate for the institution is about 80%, and the
average 6-year graduation rate is about 60%.
In this study we used five years of institutional data, which entailed 16,066
students enrolled as freshmen between (and including) the years of 2004 and 2008.
The data was collected and consolidated from various university student databases.
A brief summary of the number of records (i.e., freshman students) by year is given
in Table 2.2.
 Data Analytics Process ◾ 51

Experimental Design (10-

Fold Cross-Validation)
Raw data

Data 10%
preprocessing 10% 10%
10% 10%
Model Experiment results
10% 10%
testing (confusion matrixes)
10% 10%
Preprocessed Design of 10%
data YES NO
experiments
# of correctly # of incorrectly
Model YES predicted YES predicted YES
building
# of incorrectly # of correctly
NO
Prediction models predicted NO predicted NO

Model
deployment Sensitivity
analysis

Decision trees Artificial neural networks Variable

Importance
importance

1 100
f(z) = 90
1 + z−1 80
70
60
50 50
40 40
30
Support vector machines Logistic regression
Variable names

Figure 2.6 Analytics process employed for the student attrition prediction study.

Table 2.2 Five-year Freshmen Student Data Used in this Study

Total Number of Returned for Freshman
Year Freshmen Students the 2nd Fall Attrition (%)

2004 3249 2541 21.79%

2005 3306 2604 21.23%

2006 3234 2576 20.35%

2007 3207 2445 23.76%

2008 3070 2391 22.12%

Total: 16066 Total: 12557 Average: 21.84%

The data contained variables related to the student’s academic, financial,

and demographic characteristics. A complete list of variables obtained from the
student databases is given in Table 2.3. After converting the multidimensional
student data into a flat file (a single file with columns representing the variables and
rows representing the student records), the file was assessed and preprocessed to
identify and remove anomalies and unusable records. For instance, we removed all
52 ◾ Analytics and Knowledge Management

Table 2.3 Variables Obtained from Student Records

No Variables Data Type

1 College Multi Nominal

2 Degree Multi Nominal

3 Major Multi Nominal

4 Concentration Multi Nominal

5 Fall hours registered Number

6 Fall earned hours Number

7 Fall GPA Number

8 Fall cumulative GPA Number

9 Spring hours registered Number

10 Spring earned hours Number

11 Spring GPA Number

12 Spring cumulative GPA Number

13 Second fall registered (Y/N) Nominal

14 Ethnicity Nominal

15 Sex Binary Nominal

16 Residential code Binary Nominal

17 Marital status Binary Nominal

18 SAT high score comprehensive Number

19 SAT high score English Number

20 SAT high score reading Number

21 SAT high score math Number

22 SAT high score science Number

23 Age Number

24 High school GPA Number

25 High school graduation year and month Date

26 Starting term as new freshman Multi Nominal

(Continued)
 Data Analytics Process ◾ 53

Table 2.3 (Continued) Variables Obtained from Student Records

No Variables Data Type

27 TOEFL score Number

28 Transfer hours Number

29 CLEP earned hours Number

30 Admission type Multi Nominal

31 Permanent address state Multi Nominal

32 Received fall financial aid Binary Nominal

33 Received spring financial aid Binary Nominal

34 Fall student loan Binary Nominal

35 Fall grant/tuition waiver/scholarship Binary Nominal

36 Fall federal work study Binary Nominal

37 Spring student loan Binary Nominal

38 Spring grant/tuition waiver/scholarship Binary Nominal

39 Spring federal work study Binary Nominal

international student records from the dataset because they did not contain some
of the presumed important predictors (e.g., high school GPA and SAT scores). In
the data transformation phase, some of the variables were aggregated (e.g., “Major”
and “Concentration” variables aggregated to binary variables MajorDeclared and
ConcentrationSpecified) for better interpretation for the predictive modeling.
Additionally, some of the variables were used to derive new variables (e.g., Earned/
Registered and YearsAfterHighSchool).

EarnedHours
EarnedByRegistered =
RegisteredHours
YearsAfterHighSchool = FreshmenEnrollmentYear – HighSchoolGraduationYear

The Earned/Registered hours variable was created to have a better representation of

the students’ resiliency and determination in their first semester of the freshman
year. Intuitively, one would expect greater values for this variable to have a posi-
tive impact on retention. The YearsAfterHighSchool variable was created to measure
the impact of the time taken between high school graduation and initial college
enrollment. Intuitively, one would expect this variable to be a contributor to the
54 ◾ Analytics and Knowledge Management

prediction of attrition. These aggregations and derived variables are determined

based on a number of experiments conducted for a number of logical hypotheses.
The ones that made more common sense and the ones that led to better prediction
accuracy were kept in the final variable set.
Reflecting the population, the dependent variable (i.e., “Second Fall Registered”)
contained many more yes records (approximately 80%) than no records (approxi-
mately 20%). We experimented with the options of using and comparing the results
of the models built with the original data (biased for the yes records) versus the
well-balanced data.

Predictive Analytics Models

In this study, four popular classification methods (i.e., ANN, decision trees, SVMs,
and logistic regression) along with three ensemble techniques (i.e., bagging, bust-
ing, and information fusion) are built and compared to each other using their
predictive accuracy on the holdout samples. A large number of studies compare
data mining methods in different settings. Most of these previous studies found
machine-learning methods (e.g., ANN, SVMs, and decision trees) to be superior to
their statistical counterparts (e.g., logistic regression and discriminant analysis) in
terms of both being less constrained by assumptions and producing better predic-
tion results. Our findings in this study confirm these results. What follows are brief
descriptions of the individual and ensemble prediction models used in this study.

Artificial neural networks (ANN) are biologically inspired, analytical techniques,

capable of modeling extremely complex nonlinear functions. Formally
defined, neural networks are analytic techniques modeled after the processes
of learning in the cognitive system and the neurological functions of the
brain and capable of predicting new observations (on specific variables) from
other observations (on the same or other variables) after executing a pro-
cess of so-called learning from existing data. In this study we used a popu-
lar neural network architecture called multilayer perceptron (MLP) with a
back-propagation, supervised learning algorithm. MLP, a strong function
approximator for prediction and classification problems, is arguably the most
commonly used and well-studied ANN architecture. Hornik, Stinchcombe,
and White (1990) empirically show that given the right size and structure,
MLP is capable of learning arbitrarily complex nonlinear functions to an
arbitrary accuracy level. MLP is essentially the collection of nonlinear neu-
rons (perceptrons) organized and connected to each other in a feedforward
multilayered structure. A pictorial representation of the ANN architecture
used in this study is shown in Figure 2.7.
Decision trees are powerful classification algorithms that are becoming increas-
ingly more popular due to their intuitive explainability characteristics.
Popular decision tree algorithms include Quinlan (1986, 1993)’s ID3,
 Data Analytics Process ◾ 55

Backpropagation

∫
Socio-demographic
Predicted
∫ =
vs. Actual
Financial Returned for
∫ the second fall
semester?
Educational ∫ (Yes/No)

Other

Input Hidden Output

layer layer layer

Figure 2.7 MLP-type artificial neural network architecture used in this study.

C4.5, C5, and Breiman et al. (1984)’s classification and regression trees
(CART) and chi-squared automatic interaction detector (CHAID). In this
study, we used the C5 algorithm, which is an improved version of the C4.5
and ID3 algorithms.
Logistic regression is a generalization of linear regression. It is used primarily for
predicting binary or multiclass dependent variables. Because the response vari-
able is discrete, it cannot be modeled directly by linear regression. Therefore,
rather than predicting a point estimate of the event itself, it builds the model
to predict the odds of its occurrence. While logistic regression has been a
common statistical tool for classification problems, its restrictive assumptions
on normality and independence led to an increased use and popularity of
machine learning techniques for real-world prediction problems.
Support vector machines (SVMs) belong to a family of generalized linear mod-
els that achieves a classification or regression decision based on the value of
the linear combination of features. The mapping function in SVMs can be
either a classification function (used to categorize the data, as is the case in
this study) or a regression function (used to estimate the numerical value of
the desired output). For classification, nonlinear kernel functions are often
used to transform the input data (inherently representing highly complex
nonlinear relationships) to a high dimensional feature space in which the
56 ◾ Analytics and Knowledge Management

input data becomes more separable (i.e., linearly separable) compared to

the original input space. Then, the maximum-margin hyperplanes are con-
structed to optimally separate the classes in the training data. Two parallel
hyperplanes are constructed on each side of the hyperplane that separates
the data by maximizing the distance between the two parallel hyperplanes.
An assumption is made that the larger the margin or distance between these
parallel hyperplanes, the better the generalization error of the classifier will
be (Cristianini and Shawe-Taylor, 2000).
Ensembles and bagging (random forest): A random forest is a classifier that consists
of many decision trees and outputs the class that is the mode of the classes out-
put by individual trees. A random forest consists of a collection (ensemble) of
deceivingly simple decision trees, each capable of producing a response when
presented with a set of predictor values. A random forest has shown to run
very efficiently on large datasets with a large number of variables. The algo-
rithm for inducing a random forest was first developed by Breiman (2001).
Ensembles and boosted trees: The general idea of boosted trees is to compute a
sequence of very simple trees, where each successive tree is built for the pre-
diction residuals of the preceding tree. It learns from the previous tree, in
order to construct the succeeding one so that the misclassification of cases is
minimized. Detailed technical descriptions of this methods can be found in
Hastie et al (2001).
Ensembles and information fusion: Information fusion is the process of “intelli-
gently” combining the information (predictions in this case) provided by two
or more information sources (i.e., prediction models). While there is an ongo-
ing debate about the sophistication level of fusion methods, there is a general
consensus that fusion (combining predictions) produces more accurate and
more robust prediction results (Delen, 2015).

Sensitivity Analysis
In machine-learning algorithms, sensitivity analysis is a method for identifying
the “cause and effect” relationship between the inputs and outputs of a prediction
model (Delen et al, 2017). The fundamental idea of sensitivity analysis is that it
measures the importance of predictor variables based on the change in model-
ing performance that occurs if a predictor variable is not included in the model.
Hence, the measure of sensitivity of a specific predictor variable is the ratio of the
error of the trained model without the predictor variable to the error of the model
that includes this predictor variable. The more sensitive the network is to a particu-
lar variable, the greater the performance decrease would be in the absence of that
variable, and therefore the greater the ratio of importance. This method is often
followed in machine learning techniques to rank the variables in terms of their
importance according to the sensitivity measure defined in the following equation
(Saltelli, 2002):
 Data Analytics Process ◾ 57

Vi V ( E ( Ft X i ))
=Si =
V ( Ft ) V ( Ft )
where V(Ft) is the unconditional output variance. In the numerator, the expectation
operator E calls for an integral over X-i; that is, over all input variables but Xi, then
the variance operator V implies a further integral over X i. Variable importance is
then computed as the normalized sensitivity. Saltelli et al. (2004) showed that the
equation above is the proper measure of sensitivity to rank the predictors in order of
importance for any combination of interaction and nonorthogonality among pre-
dictors. As for the decision trees, variable importance measures were used to judge
the relative importance of each predictor variable. Variable importance ranking
uses surrogate splitting to produce a scale which is a relative importance measure
for each predictor variable included in the analysis. Further details on this proce-
dure can be found in Breiman et al. (1984).

Results
In the first set of experiments, we used the original dataset which was composed
of 16,066 records. Based on the tenfold cross-validation, the SVMs produced the
best results with an overall prediction rate of 87.23%; the decision tree came out
as the runner up with an overall prediction rate of 87.16%; followed by ANN and
logistic regression with overall prediction rates of 86.45% and 86.12% respectively
(see Table 2.4). A careful examination of these results reveals that the prediction
accuracy for the “yes” class is significantly higher than the prediction accuracy of
the no class. In fact, all four model types predicted the students who are likely to
return for the second year with better than 90% accuracy while they did poorly on
predicting the students who are likely to drop out after the freshman year with less

Table 2.4 Prediction Results for tenfold Cross-Validation with Unbalanced

Dataset
ANN(MLP) DT(C5) SVM LR

No Yes No Yes No Yes No Yes

No 1494 384 1518 304 1478 255 1438 376

Yes 1596 11142 1572 11222 1612 11271 1652 11150

SUM 3090 11526 3090 11526 3090 11526 3090 11526

Per-class 48.35% 96.67% 49.13% 97.36% 47.83% 97.79% 46.54% 96.74%

Accuracy

Overall 86.45% 87.16% 87.23% 86.12%

Accuracy
58 ◾ Analytics and Knowledge Management

than 50% accuracy. Since the prediction of the “no” class is the main purpose of
this study, less than 50% accuracy for this class was deemed unacceptable. Such a
difference in prediction accuracy of the two classes can be attributed to the skew-
ness of the original dataset (i.e., approximately 80% “yes” and approximately 20%
“no” samples). Previous studies also commented on the importance of having a
balanced dataset for building accurate prediction models for binary classification
problems (Wilson and Sharda, 1994).
In the next round of experiments, we used a well-balanced dataset where the
two classes were represented equally. In realizing this approach, we took all of the
samples from the minority class (i.e., the “no” class herein) and randomly selected
an equal number of samples from the majority class (i.e., the “yes” class herein),
and repeated this process ten times to reduce the bias of random sampling. Each of
these sampling processes resulted in a dataset of 7,018 records, of which 3,509 were
labeled as “no” and 3,509 were labeled as “yes.” Using a tenfold cross-validation
methodology, we developed and tested prediction models for all four model types.
The results of these experiments are shown in Table 2.5. Based on the hold-out
sample results, SVMs generated the best overall prediction accuracy with 81.18%,
followed by decision trees, ANN, and logistic regression with overall prediction
accuracy of 80.65%, 79.85%, and 74.26% respectively. As can be seen in the per-
class accuracy figures, the prediction models did significantly better on predicting
the “no” class with the well-balanced data than they did with the unbalanced data.
Overall, the three machine learning techniques performed significantly better than
their statistical counterpart, logistic regression.
Next, another set of experiments was conducted to assess the predictive ability
of the three ensemble models. Based on the tenfold cross-validation methodology,
the information fusion type ensemble model produced the best results with an over-
all prediction rate of 82.10%, followed by the bagging type ensembles and busting
type ensembles with overall prediction rates of 81.80% and 80.21% respectively.

Table 2.5 Prediction Results for tenfold Cross-Validation with Balanced

Dataset
ANN(MLP) DT(C5) SVM LR

No Yes No Yes No Yes No Yes

Confusion No 2309 464 2311 417 2313 386 2125 626

Matrix Yes 781 2626 779 2673 777 2704 965 2464

SUM 3090 3090 3090 3090 3090 3090 3090 3090

Per-class 74.72% 84.98% 74.79% 86.50% 74.85% 87.51% 68.77% 79.74%

Accuracy

Overall Accuracy 79.85% 80.65% 81.18% 74.26%

 Data Analytics Process ◾ 59

Table 2.6 Prediction Results for the Three Ensemble Models

Boosting (Boosted Bagging (Random Information Fusion
Trees) Forest) (Weighted Average)

No Yes No Yes No Yes

No 2242 375 2327 362 2335 351

Yes 848 2715 763 2728 755 2739

SUM 3090 3090 3090 3090 3090 3090

Per-class 72.56% 87.86% 75.31% 88.28% 75.57% 88.64%

Accuracy

Overall 80.21% 81.80% 82.10%

Accuracy

(See Table 2.6 for a complete list of results for the ensembles.) Even though the pre-
diction results are slightly better than the individual models, ensembles are known
to produce more robust prediction systems compared to a single-best prediction
model (Delen, 2015).
In addition to assessing the prediction accuracy for each model type, a sen-
sitivity analysis was conducted on the developed models to identify the relative
importance of the independent variables (i.e., the predictors). In realizing the
overall sensitivity analysis results, each of the four individual model types gener-
ated its own sensitivity measures ranking all of the independent variables in a
prioritized list. Each model type generated slightly different sensitivity rankings
of the independent variables. After collecting all four sets of sensitivity numbers,
the sensitivity numbers were normalized and aggregated into a single table (see
Table 2.7).
Using the numerical figures from Table 2.7, a horizontal bar chart is created
to pictorially illustrate the relative sensitivity or importance of the independent
variables (see Figure 2.8). In Figure 2.8, the y-axis lists the independent variables
in the order of sensitivity or importance from top (most important) to bottom (the
least important) while the x-axis shows the aggregated relative importance of each
variable.
The x-axis denotes the normalized relative importance measure for independent
variables.

Discussion and Conclusions

The results show that, given sufficient data with the proper variables, predictive
analytics methods can predict freshmen student attrition with approximately 80%
accuracy. Results also showed that, regardless of the prediction model employed,
60 ◾ Analytics and Knowledge Management

Table 2.7 Aggregated Sensitivity Analysis Results

Variable Name Ann DT SVM LR Sum

YearsAfterHighSchool 0.0020 0.0360 0.0030 0.0040 0.0450

Age 0.0085 0.0360 0.0000 0.0010 0.0455

HighSchoolGpa 0.0050 0.0360 0.0060 0.0000 0.0470

HighSchoolGraduationMonth 0.0070 0.0360 0.0030 0.0010 0.0470

StartingTerm 0.0075 0.0360 0.0000 0.0040 0.0475

Sex 0.0110 0.0360 0.0000 0.0010 0.0480

ConcentrationSpecified 0.0130 0.0360 0.0000 0.0000 0.0490

MajorDeclared 0.0065 0.0360 0.0000 0.0080 0.0505

ReceivedFallAid 0.0125 0.0360 0.0030 0.0010 0.0525

TransferredHours 0.0080 0.0360 0.0030 0.0080 0.0550

SatHighReading 0.0080 0.0360 0.0000 0.0120 0.0560

SatHighComprehensive 0.0216 0.0360 0.0000 0.0000 0.0576

SpringFederalWorkStudy 0.0220 0.0360 0.0000 0.0010 0.0590

ClepHours 0.0210 0.0360 0.0030 0.0010 0.0610

SatHighScience 0.0230 0.0360 0.0000 0.0040 0.0630

PermanentAddressState 0.0270 0.0360 0.0000 0.0010 0.0640

FallGrantTuitionWaiverScholarship 0.0280 0.0360 0.0000 0.0000 0.0640

FallFederalWorkStudy 0.0240 0.0360 0.0030 0.0080 0.0710

SatHighEnglish 0.0216 0.0360 0.0030 0.0180 0.0786

SatHighMath 0.0224 0.0360 0.0060 0.0200 0.0844

Ethnicity 0.0460 0.0385 0.0060 0.0010 0.0915

AdmissionType 0.0610 0.0360 0.0000 0.0010 0.0980

MaritalStatus 0.0800 0.0360 0.0060 0.0010 0.1230

FallStudentLoan 0.0700 0.0360 0.0200 0.0000 0.1260

FallRegisteredHours 0.0490 0.0360 0.0180 0.0300 0.1330

(Continued)
 Data Analytics Process ◾ 61

Table 2.7 (Continued) Aggregated Sensitivity Analysis Results

Variable Name Ann DT SVM LR Sum

SpringGrantTuitionWaiverScholarship 0.0605 0.0360 0.1100 0.1800 0.3865

FallGpa 0.1800 0.0550 0.1750 0.0000 0.4100

SpringStudentLoan 0.0750 0.0360 0.1500 0.1900 0.4510

EarnedByRegistered 0.1100 0.0430 0.5100 0.5100 1.1730

EarnedByRegistered
SpringStudentLoan
FallGPA
SpringGrantTuitionWaiverScholarship
FallRegisteredHours
FallStudentLoan
MaritalStatus
AddmisionType
Ethnicity
SATHighMath
SATHighEnglish
FallFederalWorkStudy
FallGrantTuitionWaiverScholarship
PermenantAddressState
SATHighScience
CLEPHours
SpringFederalWorkStudy
SATHighComprehensive
SATHighReading
TransferedHours
ReceivedFallAid
MajorDeclared
ConsentrationSpecified
Sex
StartingTerm
HighSchoolGraduationMonth
HighSchoolGPA
Age
YearsAfterHS
0.00 0.20 0.40 0.60 0.80 1.00 1.20

Figure 2.8 A graphical representation of the sensitivity analysis results.

62 ◾ Analytics and Knowledge Management

the balanced dataset (compared to the unbalanced or original dataset) produced

better prediction models for identifying the students who are likely to drop out of
college prior to their sophomore year. Among the four individual prediction models
used in this study, SVMs performed the best, followed by decision trees, neural
networks, and logistic regression. From the usability standpoint, despite the fact
that SVMs showed better prediction results, one might choose to use decision trees
because compared to SVMs and neural networks, they portray a more transpar-
ent model structure. Decision trees explicitly show the reasoning process of differ-
ent prediction outcomes, providing a justification for a specific outcome, whereas
SVMs and ANN are mathematical models that do not provide such a transparent
view of how they do what they do.
Recent trends in forecasting are leaning toward using a combination of forecast-
ing techniques (as opposed to one that performed the best based on the test dataset)
for a more accurate and more robust outcome. That is, it is a good idea to use these
three models together for predicting the freshmen students who are about to drop
out, as they confirm and complement each other. These types of models are often
called ensembles. In this study, we developed prediction models using three main
types of ensembles: bagging, busting, and information fusion. It is shown that the
prediction results of ensembles are better that those of the individual ones. As men-
tioned before, the advantage of using an ensemble is to have not only slightly better
prediction results but also a prediction system that is robust in its predictions.
Successful student retention practices at the institutional level follow a multi-
step process, which starts with determining, storing (in a database), and using stu-
dent characteristics to identify the at-risk students who are more likely to drop out,
and ends with developing effective and efficient intervention methods to retain
them. In such a process, data mining can play the critical role of accurately predict-
ing attrition as well as explaining the factors underlying the phenomenon. Because
machine learning methods (such as SVMs, ANN, and decision trees) are capable of
modeling highly nonlinear relationships, they are more appropriate techniques to
predict the complex nature of student attrition with a high level of accuracy.
The success of an analytics project relies heavily on the richness (quantity
and quality) of the data representing the phenomenon under consideration. Even
though this study used a large sample of data (covering 5 years of freshman stu-
dent records) with a rather rich set of features, more data and more variables can
potentially help improve the predictive modeling results. These variables, which are
mentioned in recent literature as important, include the student’s social interaction
(being a member of a fraternity or other social groups), the student’s prior expecta-
tion from his educational endeavors, and the student’s parents’ educational and
financial background. Once the initial value of this quantitative analysis is realized
by the institution, new and improved data collection mechanisms can be put in
place to collect and potentially improve the analysis results.
As the sensitivity analysis of the trained prediction models indicate, the most
important predictors for student attrition are those related to the past and present
 Data Analytics Process ◾ 63

educational success of the student and whether they are getting financial help. To
improve the retention rates, institutions may choose to enroll more academically
successful students, and provide them with financial assistance. Also, it might be
of interest to monitor the academic experience of freshmen students in their first
semester through looking at a combination of grade point average and the ratio of
completed hours over enrolled hours.
The focus (and perhaps the limitation) of this study is the fact that it aims to
predict attrition using institutional data. Even though it leverages the findings of
the previous theoretical studies, this study is not meant to develop a new theory;
rather, it meant to show the viability of predictive analytics methods as a means to
provide an alternative to understanding and predicting student attrition at higher
education institutions. From the practicality standpoint, an information system
encompassing these prediction models can be used as a decision aid to student
success and management departments who are serious about improving retention.
Potential future directions of this study include (1) extending the predictive
modeling methods and ensembles with more recent techniques such as rough set
analysis and meta-modeling, (2) enhancing the information sources by including
the data from survey-based institutional studies (which are intentionally crafted
and carefully administered for retention purposes) in addition to the variables in
the institutional databases, and (3) deployment of the system as a decision aid for
administrators to assess its suitability and usability in the real world.

References
Berger, J.B. and Braxton, J.M. (1998). Revising Tinto’s interactionalist theory of student
departure through theory elaboration: Examining the role of organizational attri-
butes in the persistence process, Research in Higher Education 39(2): 103–119.
Berger, J.B. and Milem, J.F. (1999). The role of student involvement and perceptions of
integration in a causal model of student persistence, Research in Higher Education
40(6): 641–664.
Breiman, L. ( 2001). Random forests, Machine Learning 45(1): 5–32.
Breiman, L., Friedman, J.H., Olshen, R.A., and Stone, C.J. (1984). Classification and
Regression Trees. Monterey, CA: Wadsworth and Brooks/Cole Advanced Books and
Software.
Cabrera, A.F., Nora, A., and Castaneda, M.A. (1993). College persistence: Structural equa-
tions modeling test of an integrated model of student retention, Journal of Higher
Education 64(2): 123–139.
Caison, A.L. (2007). Analysis of institutionally specific retention research: A comparison
between survey and institutional database methods, Research in Higher Education
48(4): 435–449.
Chapman, P., Clinton, J., Kerber, R., Khabaza, T., Reinartz, T., Shearer, C., and Wirth, R.
(2000). CRISP-DM 1.0 step-by-step data mining guide. https://www.the-modeling-
agency.com/crisp-dm.pdf (accessed January 2018).
Cristianini, N. and Shawe-Taylor, J. (2000). An Introduction to Support Vector Machines
and Other Kernel-based Learning Methods. London, UK: Cambridge University Press.
64 ◾ Analytics and Knowledge Management

Deberard, S.M., Julka, G.I., and Deana, L. (2004). Predictors of academic achievement
and retention among college freshmen: A longitudinal study, College Student Journal
38(1): 66–81.
Delen, D. (2010). A comparative analysis of machine learning techniques for student reten-
tion management, Decision Support Systems 49(4): 498–506.
Delen, D. (2011). Predicting student attrition with data mining methods, Journal of College
Student Retention: Research, Theory & Practice 13(1): 17–35.
Delen, D. (2015). Real-World Data Mining: Applied Business Analytics and Decision Making.
Upper Saddle River, NJ: FT Press.
Delen, D., Sharda, R., and Kumar, P. (2007). Movie forecast guru: A web-based DSS for
Hollywood managers, Decision Support Systems 43(4): 1151–1170.
Delen, D., Tomak, L., Topuz, K., and Eryarsoy, E. (2017). Investigating injury severity
risk factors in automobile crashes with predictive analytics and sensitivity analysis
methods. Journal of Transport & Health 4: 118–131.
Fayyad, U., Piatetsky-Shapiro, G., and Smyth, P. (1996). From data mining to knowledge
discovery in databases. AI Magazine 17(3): 37–54.
Gansemer-Topf, A.M. and Schuh, J.H. (2006). Institutional selectivity and institutional
expenditures: Examining organizational factors that contribute to retention and
graduation, Research in Higher Education 47(6): 613–642.
Hastie, T., Tibshirani, R., and Friedman, J. (2001). The Elements of Statistical Learning:
Data Mining, Inference, and Prediction. New York: Springer.
Hermaniwicz, J.C. (2003). College Attrition at American Research Universities: Comparative
Case Studies. New York: Agathon Press.
Hornik, K., Stinchcombe, M., and White, H. (1990). Universal approximation of an
unknown mapping and its derivatives using multilayer feed-forward network, Neural
Networks 3: 359–366.
Kohavi, R. (1995). A study of cross-validation and bootstrap for accuracy estimation and
model selection, in the Proceedings of the 14th International Conference on AI (IJCAI),
San Mateo, CA: Morgan Kaufmann, pp. 1137–1145.
Lemmens, A. and Croux, C. (2006). Bagging and boosting classification trees to predict
churn, Journal of Marketing Research 43(2): 276–286.
Miller, T.E. and Herreid, C.H. (2010). Analysis of variables: Predicting sophomore persis-
tence using logistic regression analysis at the University of South Florida, College and
University 85(1): 2–11.
Miller, T.E. and Tyree, T.M. (2009). Using a model that predicts individual student attri-
tion to intervene with those who are most at risk, College and University 84(3): 12–21.
Piatetsky, G. (2014). CRISP-DM, still the top methodology for analytics, data mining, or
data science projects, KDnuggets. Retrieved from http://www.kdnuggets.com/2014/10/
crisp-dm-top-methodology-analytics-data-mining-data-science-projects.html
Quinlan, J. (1986). Induction of decision trees, Machine Learning 1: 81–106.
Quinlan, J. (1993). C4.5: Programs for Machine Learning. San Mateo, CA: Morgan
Kaufmann.
Saltelli, A. (2002). Making best use of model evaluations to compute sensitivity indices,
Computer Physics Communications 145: 280–297.
Saltelli, A., Tarantola, S., Campolongo, F., and Ratto, M. (2004). Sensitivity Analysis in
Practice—A Guide to Assessing Scientific Models. Hoboken, NJ: John Wiley & Sons.
 Data Analytics Process ◾ 65

Sharda, R., Delen, D., and Turban, E. (2017). Business Intelligence, Analytics, and Data
Science: A Managerial Perspective (4th ed.). London, UK: Pearson Education.
Shearer, C. (2000). The CRISP-DM model: The new blueprint for data mining, Journal of
Data Warehousing 5: 13–22.
Thammasiri, D., Delen, D., Meesad, P., and Kasap, N. (2014). A critical assessment of
imbalanced class distribution problem: The case of predicting freshmen student attri-
tion, Expert Systems with Applications 41(2): 321–330.
Thomas, E.H. and Galambos, N. (2004). What satisfies students? Mining student opinion
data with regression and decision tree analysis, Research in Higher Education 45(3):
251–269.
Tinto, V. (1993). Leaving College: Rethinking the Causes and Cures of Student Attrition.
(2nd ed.). Chicago, IL: The University of Chicago Press.
Veenstra, C.P. (2009). A strategy for improving freshman college retention, Journal for
Quality and Participation 31(4): 19–23.
Wilson, R.L. and Sharda, R. (1994). Bankruptcy prediction using neural networks, Decision
Support Systems 11: 545–557.
Chapter 3

Transforming Knowledge
Sharing in Twitter-Based
Communities Using
Social Media Analytics
Nicholas Evangelopoulos, Shadi Shakeri,
and Andrea R. Bennett

Contents
Introduction ........................................................................................................68
Collective Knowledge within Communities of Practice .......................................69
Evolution of Analytics in Knowledge Management .............................................71
Social Media Analytics.........................................................................................75
Twitter-Based Communities as Communities of Practice ................................76
Twitter-Based Communities as Organizations ................................................ 77
Transforming Tacit Knowledge in Twitter-Based Communities ..................78
Representing Twitter-Based Community Knowledge in
a Dimensional Model .................................................................................78
User Dimension ..................................................................................................81
Interaction among Users......................................................................................85
Time Dimension .................................................................................................92
Location Dimension............................................................................................94
Topic Dimension ................................................................................................97

67
68 ◾ Analytics and Knowledge Management

Topic-Time Interaction .....................................................................................102

Opinion Dimension ..........................................................................................105
Opinion-Location Interaction ........................................................................... 110
Conclusion and the Road Ahead .......................................................................113
Summary........................................................................................................... 115
References ......................................................................................................... 116

“I suggest that collective intelligence be taken seriously as a scientific

and social goal” (Gruber, 2008, p. 5).

Introduction
In the early years of knowledge-based systems development, two influential
approaches emerged within the artificial intelligence community. The first
approach was that of expert systems, where knowledge was acquired by
engineers through copious interviews with subject-matter experts. Typically, the
engineer would spend as many as 100–200 hours performing the interviews, to
collect the raw material needed to build a knowledge base that represented the
declarative and procedural knowledge of the human experts in a format that
facilitated its retrieval (e.g., Diederich, Ruhmann, & May, 1987; Gonzalez &
Dankel, 1993; Rich, Knight, & Nair, 2009, p. 427; Sharda, Delen, & Turban,
2018, p. 13). A second approach was rule-induction systems, where algorithms
processed training sets of examples and searched automatically for patterns that
could be coded or quantified (e.g., Gonzalez & Dankel, 1993, p. 62; Rich et al.,
2009, p. 355). This approach had an inherent advantage: interaction between
knowledge engineers and human experts was no longer necessary. All that was
necessary was a good algorithm, a fast machine, and a good amount of train-
ing data. Eventually, this approach led to the explosion of machine learning,
data mining, and stock-trading bots. However, as the effort to acquire knowledge
from individual experts was side-tracked, a new opportunity became practically
possible: the acquisition of socially constructed, tacit knowledge from human
communities and organizations.
In the era of social media, Big Data, and the internet of things, social media
communities fully operate in cyberspace, their primary function is contributing to
a social media document collection (corpus), and the knowledge of their members
is inscribed in the corpus in ways that do not make it readily available without some
processing. Social media analytics (SMAs) offer the opportunity to probe a social
media corpus, extract its inscribed knowledge, and transform its tacit elements to
explicit. In this chapter, we present a systematic approach to this probing and we
demonstrate its application using vignettes that involve various Twitter-based com-
munities (TBCs).
 Transforming Knowledge Sharing ◾ 69

In many organizational environments, especially in the corporate world, the

knowledge holder tends to hoard knowledge for the purposes of increasing his or
her position in the organizational power structure (De Long & Fahey, 2000; Riege,
2005). However, as organizations understand the benefits of knowledge retention,
they feel the need to provide incentives for knowledge transfer. While in the cor-
porate domain, such a transfer may be a challenge, in the domain of social media
communities, where social media platforms provide an environment for people
to engage in knowledge sharing as a community practice, knowledge transfer is
easier to address. In this chapter, we propose a dimensional model for the design
of a data warehouse for the knowledge base, arguing that the underlying indi-
vidual tacit knowledge and the corresponding collective tacit knowledge hidden
in a user-generated social media corpus can be converted into explicit knowledge
through the use of advanced analytics techniques. While a traditional data ware-
house stores business facts that reflect what is important to a business, such as
sales dollar amounts, inventory levels, or account balances, our data warehouse for
community-based knowledge stores community knowledge facts, such as topics of
interest and opinions that reflect what is important to a community. And, while
traditional data warehouse facts are compiled by running queries on a transactional
database, our knowledge base facts are compiled by running analytics on a com-
munity corpus. After such a knowledge base is built, it can be used by community
members for a variety of purposes, such as raising awareness of a public health issue,
promoting a movement, or participating in civic engagement. It can also be used
by organizations that wish to include social media sites in their strategy for build-
ing a knowledge organization or simply for building an attractive workplace that is
mindful of organizational communities and their needs. The vignettes we present
in the next sections provide examples of the analytics that can help a community or
organization move closer to these goals.

Collective Knowledge within Communities of Practice

The advancement of information technologies has revolutionized the way individ-
uals communicate. The web has evolved from a static information-sharing envi-
ronment to a platform where users can interact, collaborate, accomplish common
goals, and collectively create knowledge. For instance, social networking sites,
such as Twitter, allow users to form communities and create community-level
knowledge through socialization, dialog, and conversation. Engaging in com-
munities is inevitable, because they not only connect individuals of similar inter-
ests and build relationships that will result in feelings of belonging and mutual
commitment (Wenger, 1998), but they also allow us to construct meaning and
negotiate it with other members of our communities. Nonaka (1991) argues that
“socialization is a rather limited form of knowledge creation” (p. 99), because
70 ◾ Analytics and Knowledge Management

interaction between members of a community allows the members to socially

construct reality and constantly reshape it through discourse. This view sub-
scribes to the constructionist perspective, which proposes that sense-making, or
meaning-making, takes place when individuals constantly construct and recon-
struct their understanding of the world through social interactions and relation-
ships (Berger & Luckmann, 1966). Thus, community-level knowledge reflects
the world-view of the whole community, as opposed to an individual member of
that community.
Learning is integral to communities. Wenger, McDermott, and Snyder (2002)
argue that “community creates the social fabric of learning” (p. 28), and the social
theory of learning suggests that learning occurs through practice, community
engagement, meaning creation, and identity development. In other words, after
developing a sense of belonging to a specific community, individuals learn through
engagement (i.e., direct experience of the world), imagination (i.e., their individual
and collective images of the world), and alignment (i.e., coordinating their local
activities with other processes) (Wenger, 1998). These three modes of belonging
determine the degree to which an individual must engage in learning (Wenger,
2000) to remain competent within a particular context. Several learning theories
emphasize the importance of context in learning. For example, the theory of situ-
ated learning posits that learning occurs in context (is situated) and is the pro-
cess of participating in a community of practice (CoP) (Lave, 1988, 1991; Lave &
Wenger, 1991).
Humans, as social creatures, belong to CoPs, which are among the basic build-
ing blocks of social learning systems (Davenport & Hall, 2002). Collaborative
activities within a CoP result in the creation of a large body of language artifacts.
Jubert (1999), in this regard, describes CoPs as “a flexible group of professionals,
informally bound by common interests, who interact through interdependent tasks
guided by a common purpose thereby embodying a store of common knowledge”
(p. 166). CoPs present a context in which individuals can interact, build relation-
ships, establish norms and mutual trust, and collectively accomplish tasks. As
technologies have advanced, CoPs have evolved beyond their traditional bound-
aries (e.g., a physical classroom or office space). Recently, online CoP platforms
have become more commonly used for collaborative work and social engagement.
As a result, concepts such as computer-supported cooperative work (CSSW) or
computer-supported cooperative learning (CSCL) have emerged in the literature
(e.g., Grudin, 1994; Palmer & Fields, 1994). Research on these concepts aims to
demonstrate the way online platforms should be designed to support CSSW and
learning. The design of such platforms may include developing user-friendly inter-
faces, addressing the social dynamics of community activities, standardizing vari-
ous processes, and coordinating the complex interactions between multiple tasks
performed by multiple user communities.
As discussed above, communicating and collaborating at a community level,
whether via physical or online CoPs, lead to the creation of collective knowledge
 Transforming Knowledge Sharing ◾ 71

(collective intelligence). Collective knowledge is created through an aggregation

of many individual user contributions through participation and collaboration
(Larsson, Bengtsson, Henriksson, & Sparks, 1998; Gruber, 2008). Collective
knowledge systems (CKSs) are social ecosystems in which human (i.e., the intel-
ligent user) and computer components interact, and such interaction leads to the
accumulation of collective knowledge. Gruber (2008) defines CKSs as machines
that collect and harvest a large amount of human-generated knowledge via the
queries built by the human user. Such systems offer capabilities, including content
sharing, cocreating, coediting, and construction of knowledge that reflect the col-
lective intelligence of the users. Intelligent users are critical to CKSs, as they not
only create content, which is later aggregated for generating collective intelligence,
but also provide intelligent feedback to the system (e.g., which queries where the
most effective at addressing the problem at hand).
Web 2.0 tools, referred to as collective intelligence tools, such as social net-
working applications, facilitate the creation of learning communities, which have
specific socio-cultural contexts, and the production of collaborative knowledge.
This knowledge is embedded in collections of artifacts generated by the commu-
nity members as they interact and collaborate. The collections can be analyzed
to glean insights that point to underlying elements of community knowledge
(Gunawardena et al., 2009). The process of analyzing these collections can generate
various intelligent solutions. The macro-micro information framework proposes
that the knowledge obtained from the analysis of social media contents can be used
for informing unintended users at both macro- (e.g., society or community) and
micro- (e.g., individual) levels. For example, opinions expressed in Twitter post-
ings are used to predict stock market value. Complex content analysis techniques
(e.g., document summarization, topic extraction, and pattern recognition, etc.) are
necessary for the analysis of the aggregated content (Evangelopoulos, Magro, &
Sidorova, 2012).

Evolution of Analytics in Knowledge Management

Traditionally, knowledge management (KM) experts have concerned themselves
with finding new methods of creating and maintaining corporate intellectual capi-
tal to meet the needs of the society (Wiig, 1997a). Effective KM processes require
knowledge infrastructure (KI). KI may consist of “technology, structure, and cul-
ture along with a process architecture of acquisition, conversion, application, and
protection” of knowledge (Gold, Malhotra, & Segars, 2001, p. 185). KI permits
organizations to create, capture, and accumulate tacit knowledge in a repository or
knowledge base. The knowledge base assists individuals within an organization in
solving or addressing emerging and future problems, making evidence-based deci-
sions and predictions, and creating innovative ideas later used for developing new
products and services.
72 ◾ Analytics and Knowledge Management

Data analytics tools and techniques, as part of KI, have improved access to
rich sources of data and have improved knowledge discovery. They have also con-
tributed to the development of knowledge bases extracted from vast amounts
of unstructured, messy data found on social media and the web. The insights
obtained from Big Data through the application of data or text analytics have
enhanced the production of innovation and helped in sustaining the organization
in the constantly changing business environment. Therefore, the emergence of
analytics offers new avenues for knowledge managers for creating, disseminating,
adopting, and applying the knowledge that was otherwise unusable. Enhanced
with analytic tools, KM can support the decision-making process within an
organization.
Figure 3.1 depicts the evolution of KM. It is around the early 2000s that
modern analytics, such as data and text mining (Tsui, 2003), enter the fold.
These tools, such as those employed in the studies featured in this chapter, enable
analysts to discern meaningful patterns and associations within data (including
words and phrases). Data and text mining tools are important for businesses seek-
ing to engage in direct marketing, implement customer-relationship management
applications, and generate business intelligence, because their outcomes can be
utilized in processes such as decision-making, content management, and match-
ing customer segments to products and services. Though Figure 3.1 begins in the
1970s, the evolution of KM is much older, emerging in early economies based on
natural resources (Wiig, 1997b), as indicated by the emphasis on and apprecia-
tion of the knowledge held and employed by master craftsmen (e.g., blacksmiths,

Karl Wiig ushers in KM. Scandia hires

first Chief Knowledge Officer. Emphasis on transferring
Knowledge databases, portals, tacit knowledge to
Modern businesses document management, and explicit knowledge.
enter the database age groupware solutions developed. KM and social web
and networks are made Consulting firms invest in KM systems. create modern,
more efficient. Microsoft and IBM reposition their constantly evolving data
ARPANET software to emphasize KM repositories

1970s 1980s 1990s 2000s 2010s

IBM enters the PC marketplace. Social networking sites and Enterprise 2.0.
The first Knowledge Management Social media analytics.
System (KMS) is developed. Individuals and culture recognized as
Tim Berners-Lee proposes the critical to knowledge creation,
framework for the World Wide Web. dissemination, application.

Figure 3.1 Evolution of knowledge management.

 Transforming Knowledge Sharing ◾ 73

masons, tailors, etc.) and members of trade guilds. This specialized emphasis on
knowledge remained largely consistent until the late twentieth century (where
the time line of Figure 3.1 begins), when the mass availability of information
technology (IT) systems meant that business leaders had more control over the
efficiency of their firms’ manufacturing, marketing, and logistics networks.
Extensive information gathering on customers and other firms, and the ability to
store this information in databases, led to business practices such as point-of-sale
analysis and total quality management. The 1980s and the 1990s saw an overall
shift in business emphasis away from products, manufacturing, and skills and
toward the use of knowledge and other intellectual capital for increasing market
share (Wiig, 1997b). This emphasis encouraged many organizations to pursue
KM strategies, including working collaboratively with customers to understand
their wants and needs.
The emergence of social networking sites (SNSs) in the early 2000s enabled
individuals to build and maintain larger social networks, including with others
whom they knew online and in person; self publish their ideas and opinions;
and collaborate with others globally to generate and manage knowledge (as evi-
denced by the coproduction of Wikipedia entries) (Hemsley & Mason, 2013).
These SNS-associated abilities mean that SNSs enable users to engage in collective
sense making, construction of meaning, and maintenance of collective knowl-
edge. The current and ongoing tools of social media analysis (SMA) allow for a
contemporary understanding of knowledge as an object, a process, and as access
to information. SMA allows for the conversion of tacit knowledge to an explicit
knowledge object that can be stored, updated, and referenced. SMA allows for the
process of tacit knowledge creation, sharing, distribution, and usage to be explic-
itly traced across social networks. And, because SMA enables the conversion of
tacit knowledge to explicit, knowledge holders can oversee who has access to this
codified knowledge.
Figure 3.2 depicts the evolution of business intelligence and analytics (BI&A)
terminology since the 1970s, when the primary focus of such applications was
the provision of structured reports that business leaders could reference during

Enterprise resource planning. Business intelligence.

On-demand static reporting. Data/text mining.
Relational DBMS. Cloud computing, saas.

Decision support systems. Executive information systems. Social network analytics.

AI/Expert systems. Dashboards, scorecards. In-memory, in-database,
Data warehousing. big data analytics.

1970s 1980s 1990s 2000s 2010s

Figure 3.2 Evolution of decision support, business intelligence, and analytics.

74 ◾ Analytics and Knowledge Management

decision-making (Sharda et al., 2018, p. 13). Though such reports—which might

be generated daily, weekly, monthly, quarterly, or annually—were useful, manag-
ers of increasingly complex organizations needed more data. This need led to the
advent of decision support systems (DSSs), which aimed to combine data and mod-
els to enable decision makers to solve unstructured problems. The early 1980s saw
these models evolve into “rule-based expert systems” (Sharda et al., 2018, p. 13),
which utilized heuristics to codify experts’ knowledge for computer processing and
warehousing. Businesses also began to capture organizational data differently in the
1980s, when enterprise resource planning (ERP) systems first entered the fray. Also
during this time, relational database management (RDBM) replaced traditional
sequential, nonstandardized data representation schemas. “These systems made it
possible to improve the capture and storage of data, as well as the relationships
between organizational data fields, while significantly reducing the replication of
information” (Sharda et al., 2018, p. 14).
With the 1990s came the need for more versatile reporting and the development
of executive information systems (EISs), graphics-based dashboards that provided
executives with visually appealing snapshots of the most important key perfor-
mance indicators (KPIs) for decision-making (Sharda et al., 2018, p. 14). Data ware-
house (DW) repositories were also developed to maintain the integrity of Business
Intelligence (BI) systems, while also enabling highly visual EIS reporting. Over a
relatively short period, most medium-to-large firms had adopted EIS and DW.
The use of “BI systems” to refer to DW-driven DSSs emerged in the 2000s
because the capabilities of these online repositories had evolved to handle the ever-
changing storage needs of modern businesses (Sharda et al., 2018, p. 14). Specific
DW evolutions were given names such as “real-time data warehousing” and “right-
time data warehousing,” which reflect vendors’ and organizations’ emphasis on the
need for fresh data for decision-making. Data and text mining became popular
in the new millennium, largely because of the advancements in DW. Specifically,
because DWs could store massive amounts of data, companies began to recognize
the benefits of mining that data for bits of institutional knowledge that could be
used to improve business practices and processes. The mining of this data, in turn,
created a need for even more data storage and processing power, problems that large
businesses had the resources to solve but left smaller firms by the wayside. The needs
of these small-to-medium firms led to the development of “service-oriented archi-
tecture and software and infrastructure-as-a-service analytics business models”
(Sharda et al., 2018, p. 14), which provided organizations with as-needed access
to analytics capabilities, whereby they were billed only for the services they used.
Among the most challenging and interesting tools available to contemporary orga-
nizations are social media and social networking (Sharda et al., 2018, p. 15). Terms
such as Big Data convey the complexity of the rich, unstructured information pro-
vided by these sites, in addition to the vast challenges inherent in the analysis of
this data.
 Transforming Knowledge Sharing ◾ 75

This chapter views a modern-day overlap in the evolutions of KM and BI&A,

a convergence that has been steadily increasing since the turn of the twenty-first
century. As depicted in Figures 3.1 and 3.2, the early 2000s brought to KM
Enterprise 2.0 and its associated social networking sites and SMAs. Also during
this time, individuals became recognized as critical to knowledge creation, dis-
semination, and application, while BI analysts began to mine DWs for the knowl-
edge created by organizational members. Today, KM’s emphasis on the conversion
of tacit to explicit knowledge means that businesses aim to store such knowl-
edge in constantly evolving DWs. All of these phenomena coincide with SMA,
as undertaken by the vignettes presented here. SMA can infiltrate Enterprise 2.0
documents, uncover their underlying tacit knowledge, and expose overarching
KPIs, including authors, times, locations, opinions, and themes, that might be
useful to decision makers.

Social Media Analytics

Today’s business climate is complex and in constant transformation. This mobil-
ity demands access to new insights for predicting the future and for making
well-informed decisions. To support such decisions, the application of complex
tools and methodologies (i.e., data analytic techniques) for deriving the collective
knowledge of the corpus is paramount. BI refers to the application of analytic
techniques to turn a business into an intelligent entity, which is capable of think-
ing, predicting, and making evidence and data-driven decisions. BI is a frame-
work of various elements (e.g., architecture, tools, databases, analytical tools, etc.)
for maintaining the sustainability and stability of the firm. SMA is part of the
traditional BI framework; though the two might be different in scope and nature.
For instance, the data used in traditional analytics is mostly structured and his-
torical, whereas social media data is messy, vast, diverse, and real-time. (Sharda
et al., 2018).
SMA is “the art and science of extracting valuable hidden insights from
vast amounts of semi-structured and unstructured social media data to enable
informed and insightful decision making” (Khan, 2015, p. 1). According to
Sharda et al. (2018), SMA is defined as the systematic and scientific methods
used for drawing “valuable insights hidden in all the user-generated content
on social media sites” (p. 311). The knowledge gained from SMA techniques
reflect both the individual and collective intelligence of a community. Sharda
et al. (2018) explain that social media data is a “collective good” (p. 4), meaning
that data on social media platforms and social technologies is user-generated,
such that social interactions among people support the creation, sharing, and
exchange of information, ideas, and opinions in virtual communities that are
equivalent to CoPs.
76 ◾ Analytics and Knowledge Management

Khan (2015) proposed a multi-dimensional SMA model described as seven lay-

ers of SMAs. The model implies that text, network, actions (e.g., like, dislike, share,
etc.), hyperlinks, mobile, location, and search engines comprise SMA. Moreover,
the model indicates that, while some of these aspects (e.g., text and actions) are
easily identifiable in data, others are hidden (e.g., location and networks) and can
only be uncovered by in-depth analytics. Similarly, Sharda et al. (2018) express
that, by categorizing social media data into distinct groupings (i.e., keywords,
content grouping, geography, time of day, and landing page), data analysts can
benefit from the full potential of analytics techniques. For instance, dividing data
based on the geographic locations of users enables analysts to identify the sources
of other data points (e.g., topics, users, actions, etc.). Another example is that of
the temporal segmentation of data, which assists in the understating of complex
phenomena (e.g., the peak hours of the day for customer service requests) and
in making data-driven decisions (e.g., the hours that customer service should be
offered).
These discussions point to the need for a data warehouse that can support KM
and SMA functions. We introduce a model for such a data warehouse later in this
chapter, arguing that TBCs are developed around certain database modeling dimen-
sions, and proposing author, time, topic, opinion (i.e., sentiment), and location as
a starting set. These dimensions represent attributes of a corpus, and, at the same
time, characteristics of the communities that generated the corpus. Furthermore,
we recommend the process of compiling these dimensions in their physical form,
as database tables, as a roadmap to the extraction of a community’s collective intel-
ligence through processing its corpus. In KM terms, the product of this process con-
stitutes the community’s collective tacit knowledge. Therefore, before we introduce
our data warehouse model we provide some additional discussion on how TBCs
build their collective tacit knowledge as they function as communities of practice
and as organizations.

Twitter-Based Communities as Communities of Practice

We argue that TBCs are CoPs because they connect groups of individuals through
shared expertise and passion for accomplishing mutual goals (Wenger & Syder,
1999). Like a CoP, a TBC is built for various reasons and can center around the
creation of knowledge across diverse dimensions (Davenport, 2001). Twitter users
form or join a TBC based on the common goals and interests of the commu-
nity, and they engage in the practice of expanding knowledge as a collaborative
activity (Gunawardena et al., 2009). Community knowledge is gradually accu-
mulated through social interaction, experience sharing, observation, informal
relationships, and mutual trust (Panahi, Watson, & Partridge, 2012), and this
collective knowledge—the collective intelligence or wisdom of the crowd—can
be employed for informing users beyond the community boundaries (i.e., at the
 Transforming Knowledge Sharing ◾ 77

broader society level). As Gruber (2008) argues, “true collective intelligence can
emerge if the data collected from all those people is aggregated and recombined”
to develop new insights and new learning that would otherwise be hard to achieve
at the individual level (p. 5). The success in the achievement of this goal depends
on the use of appropriate data analytic techniques. We present some of these
techniques later in this chapter.

Twitter-Based Communities as Organizations

Even though TBCs are CoPs, they are also organizations of individuals in which
shared meanings and actions play critical roles in shaping members’ experi-
ences. Being CoPs, TBCs involve communities of people with similar interests
engaging in the practices of meaning construction and negotiation, learning,
and producing shared repositories of collective resources (such as the TBC’s
corpus). In doing so, the community members form intimate ties and establish
norms to develop a sense of belonging, trust, and joint enterprise. The stabil-
ity of such organizations depends upon their members participating in events,
exercising opinion leadership, and contributing to the formation of membership
and other relationships. Central to all these functions is the creation of reposi-
tories of artifacts (i.e., the corpus) that constitute the fabric of social learning
(Wenger, 2000). The construction of meaning is integral to the practice aspect
of the communities and the knowledge that is collaboratively produced by their
members. Wenger (1998) asserts that “practice is about meaning as an experi-
ence of everyday life” (p. 52). Similarly, Walsh & Ungson (1991) suggest that
sense making and social functioning are essential aspects of one another. They
view organizations as networks of “intersubjectively shared meanings that are
sustained through the development and use of a common language and everyday
social interaction” (p. 60).
Sense making is necessary for creating common knowledge and coordinating
community actions based on the development of shared meanings (Weick, 1995,
p. 17). Czarniawska-Jerges (1992) claims that “shared meaning is not what is cru-
cial for collective action, but rather it is the experience of collective actions that
is shared” (p. 188). Furthermore, Burns and Stalker (1961) emphasize the impor-
tance of “social cognition” by arguing that, in working organizations, decisions
are made either as a collaborative social activity or by the knowledge provided by
others’ sense making and approval. Therefore, a TBC is a sense-making organiza-
tion as it facilitates social interactions and the formation of social cognition (i.e.,
collective intelligence), shared meaning (e.g., commonly understood language
artifacts), and shared action (e.g., collaborative decision-making). These functions
are performed through members’ participation in communal practice, as well as
the accumulation of large corpora that represent the collective knowledge of the
community.
78 ◾ Analytics and Knowledge Management

Transforming Tacit Knowledge in Twitter-Based Communities

Tacit knowledge resides in individuals’ minds. “Tacit knowledge is highly personal,”
hard to express (Nonaka, 1991, p. 98), and generated through the internalization
of explicit knowledge (i.e., captured and codified knowledge such as the contents
of journal papers, documents, etc.). Conversely, explicit knowledge is articulated
tacit knowledge. The process of converting tacit knowledge to explicit and vice
versa is explained through the knowledge spiral model, developed by Nonaka and
Takeuchi (1995), which suggests that knowledge can be placed on a spectrum from
completely tacit to completely explicit.
As individuals in TBCs collaborate, they develop implicit ways of experiencing
and learning together (Lave & Wenger, 1991). In the process of achieving shared goals,
TBC members share and merge their tacit knowledge. As a result, a collective tacit
knowledge emerges (Leonard & Sensiper, 1998; Kabir & Carayiannis, 2013). In other
words, the knowledge each person carries is necessarily complementary. Orchestra
and sports players are good examples of how the combination of individuals’ tacit
knowledge could lead to innovation beyond personal knowledge (Leonard & Sensiper,
1998). Knowledge managers believe that organizations that are more successful in
discovering and incorporating the unexpressed tacit knowledge are more capable of
designing innovative products and services (Zappavigna-Lee, 2006). Extending these
ideas (e.g., Iedema, 2003; Zappavigna-Lee & Patrick, 2004; Zappavigna-Lee et al.,
2003), we propose that TBCs encompass a kind of collective tacit knowledge that,
from the point of view of the knowledge engineer (or the analyst), cannot be acquired
at the level of isolated, individual members. In this chapter, we discuss how one can
begin to uncover collective tacit knowledge by performing analytics at the corpus level.

Representing Twitter-Based Community

Knowledge in a Dimensional Model
Document management systems (DMSs) have a long history in organizations
where a significant part of operations is based on collections of documents. Such
organizations include not only publishers or digital libraries, but also educational
institutions, insurance firms, governmental agencies, and so on. A DMS typi-
cally includes scanning, storage, and workflow routing functions (Sullivan, 2001,
p. 130). As DMSs move into the realm of KM, it becomes necessary for them
to adopt a document warehouse model with a well-designed architecture. This is
where metadata repositories become essential elements for the organization of a
document collection. In general, metadata can be broken down into four categories
(Sullivan, 2001, pp. 144–150):

◾ Document content metadata, such as author, title, subject/keywords,

description/abstract, publication source, publisher, document type, format,
language, publication date, edition, etc.
 Transforming Knowledge Sharing ◾ 79

◾ Search and retrieval metadata, such as search keys and indices, document
source, versioning, URL, source type, waiting time, username, password,
directory, search engine, etc.
◾ Text mining metadata, such as keywords, topics, clusters, other features,
summaries, opinion categories, etc.
◾ Storage metadata, such as store document index, summary index, URL
index, pathname, translation information, etc.

SNS such as Facebook, Twitter, LinkedIn, Snapchat, and so on, as computer-

mediated environments that facilitate the creation and sharing of information, base
a significant part of their operations around documents and other forms of media.
However, because SNS are typically open and community-based, their knowledge
base is not managed as formally as that of a corporate organization. Since this type
of knowledge is built around a virtual CoP, where members interact, share experi-
ences, and learn through transactions with a corpus, we refer to the community
as a corpus-based community, and its knowledge as corpus-based knowledge. The
rest of this chapter focuses on Twitter, where a TBC is treated as a special type of a
corpus-based community.
As mentioned earlier, within a TBC, as within any knowledge organization,
knowledge consists of explicit and tacit elements. How does one begin to represent
the architecture of a data warehouse that supports KM in a TBC? Figure 3.3
presents a dimensional model that includes some explicit elements. This overall
layout, also known as the star schema (e.g., Adamson & Venerable, 1998, pp. 9–17;
Kimball & Ross, 2002, p. 16), has a fact table in the middle, surrounded by

Author
Time
author_key
time_key
author_id
Corpus_fact timestamp
name
hour
screen_name author_key day
address time_key date
contact_info location_key month
category text year
favorited
favorite_count
Location reply_to_SN
status_source
location_key
retweet_count
longitude
latitude
time_zone
sector
area

Figure 3.3 Dimensional model for a Twitter corpus.

80 ◾ Analytics and Knowledge Management

dimension tables. The fact table records each Tweet with references to who is the
author, when it was posted, and from which location. Each reference is expanded
as an entry in a dedicated table, or dimension, where details for the referenced
authors, locations, and time periods are provided. This dimensional model pro-
vides a platform that facilitates the efficient generation of reports and visualiza-
tions that constitute a well-established mechanism for providing organizational
intelligence.
Occasionally, custom fact tables and custom dimensions are included in the
dimensional model (Adamson & Venerable, 1998, p, 20). As we will show later in
this chapter, our additional facts and dimensions are produced with the use of social
analytics. Such knowledge might be derived days or months after the recording of
the original core facts, (or, in the case of real-time analytics, ex post, that is, a frac-
tion of a second after the core facts are captured). To present the model, we employ
database management terminology to refer to new facts as derived facts and their
corresponding dimensions as derived dimensions. Figure 3.4 shows the process of
producing these derived facts and dimensions.
The process shown in Figure 3.4 starts with the extraction of topics or opin-
ions expressed in Tweets, using text analytics. In a preliminary phase, the results
of these analytics are stored as Tweet metadata, or, in database modeling terms,
as derived attributes of the fact table. Subsequently, data warehouse extract,
transform, and load (ETL) operations are used to organize topics and opinions
in separate tables, so that they can be managed independently and serve as a
standardized, common reference across the entire data warehouse. As shown in
Figure 3.5, the data warehouse design now looks like an expanded star schema
that includes derived (i.e., computed using analytics or database queries) facts, as
well as derived dimensions. We will refer to this configuration as a knowledge base
data warehouse (KBDW) schema.

Corpus_fact Topic
Corpus_fact author_key topic_key
Corpus_fact time_key
author_key label
author_key location_key topic_type
time_key
time_key text grain_level
location_key
location_key favorited
text
text favorite_count
favorited
favorited reply_to_SN
favorite_count Opinion
favorite_count status_source
reply_to_SN
reply_to_SN retweet_count opinion_key
status_source
status_source retweet_count topic_key opinion_type
retweet_count topic_strength
tweet_topic
opinion_key
tweet_opinion
opinion_strength

Figure 3.4 Process of producing derived facts and dimensions.

 Transforming Knowledge Sharing ◾ 81

Time
Author
time_key
author_key Corpus_fact
timestamp
author_id hour
author_key
name day
time_key
screen_name date
location_key
address month
text
contact_info year
favorited
category
favorite_count Topic
reply_to_SN
status_source topic_key
Location retweet_count label
location_key topic_type
topic_key
longitude grain_level
topic_strength
latitude opinion_key
time_zone opinion_strength Opinion
sector
area opinion_key
opinion_type

Figure 3.5 Expanded dimensional model with derived dimensions and their
interactions.

Our proposed dimensional model for a KBDW offers the opportunity to create
a knowledge base that can measure the community member contributions, identify
time trends, locate the distribution of events and information across geographical
or other types of places, and track the diffusion of sentiments and ideas across time,
space, and communities. On a conceptual level, our model bears a lot of similarity
with the analytic variant of a customer relationship management data warehouse
(Kimball & Ross, 2002, p. 144). However, the focus of this chapter is on a corpus
that reflects the activity, the accomplishments, and the knowledge of a CoP, the
analytics that can uncover this knowledge, and the database models that can orga-
nize this knowledge for easy access, retention, and transfer outside the minds of
the most entrenched community members. In the sections that follow, we examine
various dimensions of the KBDW and their interactions. We begin with the user
dimension, presented in the next section.

User Dimension
The author of a document is arguably its most important attribute. On a collec-
tive level, multiple authors produce the user dimension of a corpus. Understanding
the authors includes an investigation of who they are as community members or
organizational stakeholders. Authors can vary significantly by level of involvement
and document authoring activity. Vignette 3.1 presents an example of how the user
dimension contributes to the knowledge base of a specific TBC.
82 ◾ Analytics and Knowledge Management

VIGNETTE 3.1 Top Users Posting

Zika Tweets in April 2016
Zika is a Flavivirus-borne disease primarily communicated to
humans through infected bites of the mosquito species Aedes
aegypti. The disease also spreads through blood transfusion,
sex, and from a pregnant woman to her fetus (CDC, 2017). The
virus was initially isolated in the Zika forest of Uganda in a rhe-
sus macaque monkey in 1947. While scientists have known the
Zika virus for seven decades, it only recently became a public
health crisis, first in the Americas and then internationally. Of
the reported cases of Zika virus, about 80% are asymptomatic
(Duffy et al., 2009), but in a few people, mild symptoms such
as headaches, rash, joint pain, conjunctivitis, fever, and muscle
pain are reported. As Zika reached epidemic levels, with a
prevalence rate of 99.7 cases per 100,000 people in different
parts of the world, including the United States, 1246 new cases
were reported in 2015 alone.
The Zika virus turned into a public health crisis when it
became associated with birth defects. Medical researchers claim
that Zika infections during pregnancy can lead to various condi-
tions for fetuses and infants, including a severe birth defect called
microcephaly (CDC, 2017). The state of emergency caused by
the Zika virus lasted for a year. However, after Zika cases reached
a 95% reduction in April 2017, compared to the same month in
2016, the Brazilian government announced an end to the Zika cri-
sis (The Atlantic, 2017). The World Health Organization (WHO),
however, emphasizes that like all mosquito-borne diseases, Zika
is a seasonal disease and may return to countries where Aedes
aegypti mosquitoes commonly live.
Tweets about Zika were downloaded from the Twitter user
interface on the web using search terms that specified Zika as
a keyword and April 1–30, 2016 as the time interval. A set of
17,422 Tweets was obtained. These were produced by 7503
unique users. Table 3.1 lists the top 10 users who posted the
highest number of Tweets in the data set.
Figure 3.6 shows the cumulative Tweet percentages by
cumulative user rank. For example, the top 20% of Twitter
users who posted messages about Zika in April 2016 contrib-
uted 60.98% of all Tweets about Zika that month, while the top
80% of users contributed 91.38%. The relationship between
user rank and Tweeting frequency appears to follow a power
law. Figure 3.7 shows a graph of this relationship for the top
1000 rank-ordered users.
Table 3.1 Top 10 Users Posting Zika Tweets in April 2016
Rank User Handle (Name) User Type and Description Frequency Cumulative %

1 Hniman (Henry L. Niman) Individual; tracks and analyzes infectious diseases 340 1.95
(e.g., H1N1, Zika, etc.).

2 Zika_News (Zika News) Organization; detects and shares News about Zika 292 3.63
virus.

3 TheFlaviviruses (Flavivirus Virus) Organization; detects and share news about 227 4.93
diseases caused by the Flaviviridae family.

4 Greg_folkers (Greg Folkers) Individual; reports on infamous diseases. 205 6.11

5 AdamNeira (WorldPeace2050) Individual; Tweets on a variety of topics, including 181 7.15

disease pandemics, terrorism, etc.

6 MackayIM (Ian M. Mackay) Individual; virologist who provides medical advice. 151 8.01

7 ironorehopper (ironorehopper) Individual; (no information about the author was 136 8.79
available).

8 Crof (Crawford Kilian) Individual; Retired college teacher & writer. 127 9.52

9 Halt_Zika (#Halt_Zika) Organization; affiliated with a company working on 92 10.05

anti-Zika drugs.

10 CDCDirector (Dr. Anne Schuchat) Individual; The CDC director. 90 10.57

Transforming Knowledge Sharing ◾
83
84 ◾ Analytics and Knowledge Management

100
90
80
70
60
Tweet %

50
40
30
20
10
0
0.00 20.00 40.00 60.00 80.00 100.00
User %

Figure 3.6 Cumulative Tweet percentages by cumulative user rank.

400
350
300
Tweet frequency

250
200
150
100
50
0
1
32
63
94
125
156
187
218
249
280
311
342
373
404
435
466
497
528
559
590
621
652
683
714
745
776
807
838
869
900
931
962
993

User rank

Figure 3.7 Power law in user contributions.

A careful examination of the list of the top contributors

presented in Table 3.1. Based on their profiles, they are all
interested in the Zika virus disease. Some of the accounts
are personal and others are non-personal (group accounts).
For example, Henry L. Niman (screen name: hniman) who
posted 340 Tweets during April 2016 and ranks at the top
for that month, is a medical doctor who tracks and analyzes
infectious diseases such as H1N1 and Zika. Similarly, the Zika
News (screen name: Zika_News, 292 Tweets posted in April
2016) and Flavivirus Virus (screen name: TheFlaviviruses, 227
Tweets posted in April 2016), ranked No. 2 and 3, respec-
tively. Both are non-personal Twitter accounts, dedicated
to the detection and distribution of Zika-related news or
 Transforming Knowledge Sharing ◾ 85

User/Author
user_key Corpus_fact
user_id
name user_key
screen_name time_key
address location_key
contact_info text
category favorited
favorite_count
user_rank reply_to_ScrNm
user_freq status_source
user_cumulat% retweet_count

Figure 3.8 The user dimension in the data warehouse model, with added derived
attributes.

information. A third category of users are non-human (bots,

etc.). We assume that at the current phase of AI evolution
all non-human users act as representatives of corresponding
humans or human organizations. In the future, it may become
possible to have completely autonomous AI agents who act
on their own interests.
The analysis and the output presented here helps uncover
some of the collective knowledge about who the users are,
typically held by the core members of this community (the
Zika TBC). The data on the most active users described in
this vignette allows researchers to determine the top con-
tributors to the TBC. This is a form of tacit knowledge that
we are able to uncover through an analysis of the entire
corpus. Though we can obtain the knowledge represented
by Table 3.1 using spreadsheet manipulations on the entire
corpus, a more efficient approach is to run structured query
language (SQL) queries on a data warehouse. Figure 3.8 shows
the logical design of a small segment of such a KBDW as
applicable to Vignette 3.1: The derived attributes user_rank,
user_ frequency, and user_cumulative% in the User/Author
dimension. Table 3.1 can now be generated by running a
straightforward SQL query.

Interaction among Users

Social networks refer to a social structure composed of groups of individuals or
organizations, linked to one another with some connections or relationships (Sharda
et al., 2018, p. 304). Based on this definition, social network analysis (SNA) pro-
vides techniques and tools for investigating social structures by measuring network
86 ◾ Analytics and Knowledge Management

attributes, including centrality measures, density, clustering coefficient, distance,

and so on. Although SNA concentrates on methodology more than theory, it
enables a comprehensive understanding of users’ social activities and relationships
both at individual and collective levels. For instance, using SNA techniques, we
can learn which users manifest the properties of opinion leaders (i.e., authorities
and influential users) by studying their communication patterns. Research suggests
that such individuals with high connectivity and issue involvement are successful
in affecting the perceptions and attitudes of the rest of the network (e.g., Ibarra &
Andrews, 1993; Xu et al., 2014).
One common characteristic of social networks is clustering or network transi-
tivity (Mishra et al., 2007). Clustering refers to the property of two users (nodes)
that are neighbors with a third user, to have an increased probability of being
related (friends). In statistical terms, the clustering coefficient measures “the prob-
ability that two of one’s friends are friends themselves,” which is based on triples of
nodes (i.e., three connected nodes). The global clustering coefficient is the number
of closed triples over the total number of triples (both closed and open) in the
graph. In a graph where all nodes are entirely connected to one another, the coef-
ficient equals 1. However, in most real-world networks (e.g., social networks), the
coefficient would range between 0.1 and 0.5. A local clustering coefficient, on the
other hand, determines how close a particular node and its neighbors are to form-
ing a “closed triple” (Girvan & Newman, 2002, p. 7821). Clusters or communities
emerge as a subset of highly (fully) connected individuals that are loosely related
to other users in a network. For a cluster to form, each user or node should have
more connections within the community than with the rest of the graph (Radicchi
et al., 2004). The communities typically appear as dark, dense spots in a graph
representation.
Based on the above discussion, we argue that clusters may form closed social
circles (i.e. cliques) where each user is directly linked to every other user in
the community. Because cliques are a locked network, their members may be
exposed to redundant information. Therefore, cliqued users need to connect
to the rest of the network through weak ties to access nonredundant sources of
information. A weak tie is a user (node or vertex) that links users from across
various communities. The theory of the strength of weak ties suggests that the
probability of weak ties to act as bridges between communities is greater than
that of strong ties (Granovetter, 1983). Because trusted weak ties play a critical
role in providing access to non-redundant information, they contribute even
more positively to the transfer and receipt of useful knowledge than strong ties
(Levin & Cross, 2004).
To detect communities in a social network structure, one needs to apply tech-
niques and tools developed for performing such tasks. As a result, researchers across
various domains (e.g., applied mathematics and computer science) have advanced
algorithms and standards to undertake such duties. Traditionally, community
 Transforming Knowledge Sharing ◾ 87

detection algorithms (CDAs) compute the edge weight for every pair of users in
the network graph and create a weight index. The edge weights are the calculated
edge betweenness measures (i.e., the number of shortest paths between pairs of
vertices), adopted from Freeman (1978)’s vertex betweenness centrality. Then, the
edges are added to a network of n vertices with no edges between them in order of
their weights starting with the strongest weights to the weakest (weighted index).
As a result, communities begin to appear, as users with the most powerful network
ties are linked to one another. Unlike the traditional community detection model,
Girvan and Newman (2002) propose a method that can also determine the struc-
ture of the communities using a different clustering approach. The method focuses
on identifying of the community peripheries (using the edge betweenness), rather
than the ones formed between highly-connected individuals (the individuals with
high betweenness centralities).
As discussed earlier, CDAs compute important network features and detect
communities. As the algorithms have different capabilities in handling networks of
different sizes and complexities, and are thus utilized for different purposes, their
performances have to be tested and evaluated by through the use of benchmarks or
benchmark graphs. The GN (for Girvan and Newman) benchmark, for example,
identifies communities within very small-sized graphs by computing only specific
network features (e.g., distribution, community sizes, etc.) Although most algo-
rithms perform well on the GN benchmark graphs, they might not produce good
results for extremely large, heterogamous networks with overlapping communities
(Yang, Algesheimer, & Tessone, 2016).
Alternatively, the LFR benchmark, which stands for Lancichinetti,
Fortunato, & Radicchi, produces artificial networks to enhance the performance
of CDAs in detecting communities in large, complex networks (Lancichinetti,
Fortunato, & Radicchi, 2008; Lancichinetti & Fortunato, 2009). For instance,
Yang, Algesheimer, and Tessone (2016) employ LFR benchmark graphs to
examine and compare the accuracy and computing time of eight CDAs (i.e.,
edge betweenness, fastgreedy, infomap, label propagation, leading eigenvector,
multilevel, spinglass, and walktrap) available in the R package igraph. The study
results indicate that the multilevel algorithm was superior to all the other algo-
rithms on the LFR benchmark used for the analyses. When the algorithms were
strictly tested for accuracy, (computing time is irrelevant for computing com-
munities in small networks), infomap, label propagation, multilevel, walktrap,
spinglass, and edge betweenness algorithms outperformed the others. However,
for large networks, where computing time is a critical criterion in selecting an
algorithm, infomap, label propagation, multilevel, and walktrap were revealed
to be superior options.
In Vignette 3.2, we present an example of identified communities within
the network of users featured in Vignette 3.1 using the leading eigenvector
algorithm.
88 ◾ Analytics and Knowledge Management

VIGNETTE 3.2 Identifying User

Communities Within the Zika Network
We continue the scenario in Vignette 3.1, addressing Tweets
about Zika, a Flavivirus-borne disease primarily communi-
cated to humans through infected bites of the mosquito spe-
cies Aedes aegypti. The analysis of the network graph formed
by users’ communications about the Zika virus in April 2016
(see Figures 3.9 and 3.10) sheds light on their communication
patterns. Figure 3.9 depicts the users (i.e., who interacts) and
the intensity of their interactions (i.e., how much information is
transferred between the pair) in April 2016. For instance, the
CDC is a recognized health organization that has developed
a large number of connections (mostly in-degree) as it distrib-
utes news and information on the status of the Zika virus. The
width of the links (edges) in the network represents the number
of times the two users exchanged information between them
through replying, mentioning, and Tweeting interactions. To
better explain why each user is placed in a particular spot in
the network, we calculated the betweenness and degree cen-
tralities of each node. Degree centrality refers to the number of
connections each node has, whereas betweenness centrality
denotes the number times each node lies in geodesic paths
between other pairs of users. Our interpretation of the central-
ity measures is as follows:

Figure 3.9 Network of users within the Zika social network in April 2016.
 Transforming Knowledge Sharing ◾ 89

Figure 3.10 Communities of users within the Zika social network in April 2016.

The investigation of the degree centrality measures of

users suggests that we can categorize network users into
two distinct groupings: information authorities and informa-
tion disseminators. The information authorities are the user
accounts (e.g., @CDCgov, @WHO, @CDCDirector), which
are highly cited as credible sources of information by other
network users when disseminating information. These users,
as a result, have developed many in-degree connections
over time. The information disseminators (e.g., @ironore-
hoper, @bloglovin), on the other hand, are the nodes with
many out-degree links, because they distribute information
they receive from authoritative outlets. We observe that the
connections of users (i.e., the degree centrality) impact their
positioning in the network. For example, the @CDCgov’s
many in-degree ties have positioned it almost at the heart of
the network. Whereas, @bloglovin with only one out-degree
connection is pulled toward the margins. Similarly, calcu-
lating the vertex betweenness centrality (as proposed by
Freeman, 1978) reveals that the network has an overall low
90 ◾ Analytics and Knowledge Management

betweenness centrality. We hypothesize that this is occurring

because most nodes have a betweenness of zero (i.e., these
users are not connecting any user and community across the
graph), with only a few having betweenness of more than
10 (e.g., @Zika_News, @PeterHotez, @HelenBranswell).
Remarkably, while both degree and betweenness centrality
influence the positioning of vertices in networks, nodes with
high betweenness centrality might not necessarily have high
degree centrality. This phenomenon agrees with the theory
of the strength of weak ties, as it suggests that weak ties
(i.e., information bridges) typically have a high betweenness
with very few links (Granovetter, 1973). A good example of
this type of user in this following network is the @CDCgov,
which has a zero betweenness centrality with a relatively
high degree centrality (i.e., 17).
Figure 3.10 presents the clusters (communities) of users in
the graph. As defined earlier, network clusters are groups of
individuals with dense internal interactions and sparse exter-
nal friendships (Mishra et al., 2007). To identify the clusters,
we employed various community cluster detection algorithms
suggested by Yang, Algesheimer, and Tessone (2016), includ-
ing edge betweenness, fastgreedy, infomap, label propaga-
tion, leading eigenvector, multilevel, spinglass, and walktrap,
all implemented in the R package igraph. After obtaining the
clustering results, we observed that the leading eigenvec-
tor algorithm outperformed the others. Consequently, upon
using the leading eigenvector algorithm, we identified six
network clusters (see Table 3.2). After analyzing the com-
position of each cluster, we observed that each community
consists of a mixture of various stakeholder groups, (e.g.,
healthcare, press, politics, etc.). This observation is consis-
tent with the stakeholder view of organizations, which posits
that multiple groups of individuals are needed to influence
how a CoP achieves its goals. For instance, if the commu-
nity has the goal of raising awareness within the society at
large by providing the most accurate, detailed information
about a disease epidemic, the contribution of different types
of members (stakeholders) might be necessary to increase
the quality of the provided information, thus facilitating the
achievement of the goal. The contribution of different stake-
holders might increase the credibility of, and trust in, the
shared information.
Figure 3.11 shows the logical design of a small segment of
a KBDW that is applicable to Vignette 3.2. After obtaining the
 Transforming Knowledge Sharing ◾ 91

Table 3.2 Network Clusters for Zika Tweets in April 2016

Rank Member Count Member User Handles

1 12 AdamNeira, ushahmd, AFHSBPAGE, US_FDA, npr,

CDCgov, Contagion_Live, PeterHotez, bbbrieger,
greg_folkers, FluTrackers

2 11 CNN, nbcnews, epstein_dan, Reuters,

PAHOFoundation, statnews, ComptonCreekMAD,
nytimes, FoxNews, ZikaECN, WSJ, washingtonpost

3 7 Zika_News, AP, Laurie_Garrett, MiamiHerald,

Halt_Zika, HHSGov, SpeakerRyan

4 6 CDCDirector, WhiteHouse, NIH, DonnaYoungDC,

DrFriedenCDC, POTUS

5 6 Crof, MackayIM, hniman, ChannelNewsAsia, WHO,

HelenBranswell

6 4 Intrexon, USATODAY, Oxitec, ABC

User/Author
user_key Corpus_fact
Cluster user_id
name user_key
cluster_key screen_name time_key
cluster_rank address location_key
cluster_size contact_info text
category favorited
favorite_count
user_rank reply_to_ScrNm
user_freq status_source
user_cumulat% retweet_count
cluster_key

Figure 3.11 The derived cluster dimension in the data warehouse model.

size and the rank of each cluster of users through SNA, clus-
ters are now standing on their own, represented as separate
derived dimensions. Using a data warehouse with this configu-
ration, we could generate Table 3.2 by running a straightfor-
ward SQL query that draws information from all user clusters
and their associated users.
92 ◾ Analytics and Knowledge Management

Time Dimension
The time at which a document is created not only communicates the author’s
thoughts at that specific date, hour, and minute, but also provides clues to the
environmental context on which he or she is commenting. The time period in
which documents in a collection were generated comprises the time dimen-
sion of a corpus. Examples of time-stamped documents include news stories
(which are “news” only with reference to a specific point in time), emails and
other communication documents, server logs, customer service calls, and social
media postings.
Once the time dimension is established, various corpus statistics can be
tracked over time, forming a data structure known as a time series. These include
counts, averages, ranges, and so on, that can be easily obtained by executing data-
base queries. Time series models include regression with linear and polynomial
time trends, auto-regressive models, exponential smoothing models, time series
decomposition, and autoregressive-integrated-moving average (ARIMA) models.
For more details on these time series analytics, refer to a standard time series
forecasting text, such as Bowerman, O’Connell, & Koehler (2005). Vignette 3.3
presents an example of how the time dimension contributes to the knowledge
base of a TBC.

VIGNETTE 3.3 Tweets on

International Women’s Day 2017
International Women’s Day (IWD) is held globally on March 8
of each year as a celebration of the political and social achieve-
ments of women, while also aiming to bring attention to
instances of gender-based discrimination and call for their rec-
tification (University of Chicago, 2014; Sanghani, 2017; United
Nations, 2017). IWD is not affiliated with a single group; par-
ticipants include members of governments, women’s organi-
zations, corporations, and charities, in addition to individual
citizens (Sanghani, 2017).
After the success of the Women’s March on Washington
during the January 2017 inauguration of U.S. President Donald
Trump—an event that The Washington Post called “likely the
largest single-day demonstration in recorded U.S. history”
(Chenoweth & Pressman, 2017) —event organizers coordi-
nated an event called A Day Without Women to coincide with
IWD 2017 (Sanghani, 2017). Utilizing social media and other
outlets, the organizers encouraged women to participate in a
worldwide labor strike (Sanghani, 2017; Zarya, 2017; Pruitt,
 Transforming Knowledge Sharing ◾ 93

2017), to refrain from shopping, and to wear red in solidar-

ity with the movement (Sanghani, 2017; Pruitt, 2017). These
specific calls for participation, in addition to the tumultuous
global environment—as many citizens were still shocked at
the United Kingdom’s “Brexit” from the European Union and
Donald Trump’s victory in the U.S. presidential election—set
the scene for massive participation and demonstrations during
IWD 2017 (Sanghani, 2017).
Tweets for IWD 2017 were collected via the
hashtags #IWD, #IWD2017, #InternationalWomensDay, and
#DayWithoutAWoman using the R package twitteR and
Twitter’s API interface. An hourly trend analysis revealed that at
the height of the day—between noon and 5 p.m.—traffic would
likely reach about 70,000 Tweets per hour; thus, a formula was
generated to determine the number of Tweets that should be
collected each hour to generate a representative sample. The
Tweets collected for March 8, 2017, totaled 98,900, represent-
ing a 10% sample, stratified over time. Subsequently, the Tweets
underwent additional cleaning and formatting processes, with
the final sample of complete Tweets totaling 82,857.
The time dimension provides a snapshot of knowledge
generated within specified parameters. Researchers seeking to
conduct longitudinal studies in KM could trace patterns, congru-
ences, and incongruences among information disseminated over
time using such data. The time dimension also offers research-
ers the ability to potentially pinpoint when a specific piece of
knowledge enters the corpus, in addition to the ability to trace
the rate of its adoption into the knowledge base, as indicated
by its being shared by various users over time. Practitioners,
such as the organizers of the Day Without Women strike, can
utilize the time dimension to determine the ideal timeframe for
introducing an event or idea to the corpus, such that the optimal
level of digital word-of-mouth marketing is generated.
Figure 3.12 shows the distribution of Tweets about IWD
2017 over the 24-hour course of March 8, 2017. As would be
expected, Twitter activity climbs from the early morning to
midday and is highest between 1 p.m. and 5 p.m., which is
when most of the rallies and events were taking place in the
United States.
Figure 3.13 shows the logical design of a small segment of
a KBDW that applies to Vignette 3.3. We assume that the time
grain is set at the hour level, which means that there is unique
timekey for each hour of the day and all hours are listed in the
timetable where the hourly frequency is added as a derived
94 ◾ Analytics and Knowledge Management

80000

70000

60000
Tweet volume

50000

40000

30000

20000

10000

0
0 5 10 15 20 25
Hour of the day

Figure 3.12 Tweet volume by hour of the day (1 = 1:00 AM, 13 = 1:00 PM, etc.).

Time

time_key Corpus_fact
date
user_key
month
time_key
date
location_key
year
text
hour
favorited
hourly_frequency favorite_count
reply_to_ScrNm
status_source
retweet_count

Figure 3.13 The time dimension in the data warehouse model.

attribute. Using a data warehouse with this configuration, we

could generate the information depicted in Figure 3.12 by run-
ning a straightforward SQL query that draws information from
the timetable.

Location Dimension
The proliferation of geographic information systems (GIS) has brought attention to
the spatial dimension of various processes in the social, political, financial, and sci-
entific domains. Efficient use of spatial data, including users’ geographic location,
place of origin, place of destination, or place of interest in general, help uncover and
add to the knowledge base the location dimension, which can often be hidden in a
 Transforming Knowledge Sharing ◾ 95

corpus of collective tacit knowledge. The location at which a document is created

provides not only situational and contextual clues but also allows for understanding
the patterns of occurrence of certain events. Through tracking of the geographic
locations of events, the stakeholder can proactively address current, emerging, and
potential problems, and explain certain behaviors associated with a social event,
such as a social movement or a disease outbreak. The combination of the loca-
tion dimension with other documents’ dimensions (e.g., authors, opinion, topics,
or time) enable further decoding and extracting other hidden knowledge layers.
Vignette 3.4 presents an example of how the location dimension contributes to the
knowledge base of a TBC.

VIGNETTE 3.4 Zika Tweets in May and June 2017

We continue the scenario from Vignettes 3.1 and 3.2, address-
ing Tweets about Zika, a Flavivirus-borne disease primarily
communicated to humans through infected bites of the mos-
quito species Aedes aegypti. The Zika virus grew into a public
health crisis in 2015 and 2016, which demanded the devel-
opment of preventive and predictive measures to control the
spread of the disease. In May 2017, due to a 95% decrease in
the number of newly infected individuals, the Brazilian govern-
ment declared an end to the Zika virus emergency (BBC News,
2017). Even though the Zika crisis is over, health authorities
are concerned about the potential of another Zika emergency;
thus, asking for remaining vigilant about the disease status and
dynamics. This might entail new insights provided by the anal-
ysis of various sources of data on the web and social media.
Several researchers (e.g., Bodnar & Salathé, 2013; Carneiro &
Mylonakis, 2009; Culotta, 2010; Salathé & Khandelwal, 2011)
have employed Twitter data to improve their understanding of
predicting disease occurrences and dynamics. Twitter data is
a rich source for extracting the collective knowledge of TBCs,
and, as such, this vignette presents mining of Twitter spatial
data (i.e., Tweets’ latitudes and longitudes), a form of collective
tacit knowledge within a TBC.
In August 2009, Twitter began supporting the per-Tweet
geotagging functionality. When this feature is activated by the
user, location coordinates (i.e., latitude and longitude) are asso-
ciated with the Tweet. This enables the tracking of the Tweet
author’s location. The spatial data is findable through the Twitter
API, unless the authors opt to delete past location data from
displaying in their Tweets. Since this feature is off by default,
only a small portion of Tweets have the geo-location tag.
96 ◾ Analytics and Knowledge Management

To access the collective knowledge hidden in the Twitter

geographical data produced by the Zika community, we col-
lected the Zika Tweets posted in the months of May and June
2017 using the Zika hashtag as a query keyword and Twitter
API. In the data cleaning process, we removed Tweets lack-
ing the latitude and longitude tags. As a result, we obtained a
sample of 200 data points, which was equal to roughly 0.1% of
all the collected Tweets.
Visualization of geo-location data and spatial analytics can
be done via various packages, such as the commercial GIS
package ArcGIS, that can match each data point (latitude and
longitude) on a map. Figure 3.14 depicts the distribution of
the location data of the 200 geo-located Zika Tweets across
a world map. A similar visualization can be produced using
Tableau®, a commercial analytics and visualization package.
As observed in the visualization, most Tweets were created by
individuals in the United States and South America (especially
Brazil, the center of the epidemic), while some originate in
South Africa, various European locations, and India. We postu-
late that the Tweets’ geotags represent the countries that have
been most affected by the infection, thus actively involved in
the management of the crisis (e.g., the United States, Brazil,
and India). India is a good example to support our hypoth-
esis, as we believe that the Tweets originating in India con-
tain information about the three laboratory-confirmed cases of
Zika virus disease reported in the Bapunagar area, Ahmedabad
district, and Gujarat state (The New York Times, 2017; WHO,
2017). These Tweets might carry important information about

Figure 3.14 Dispersion of Tweets about Zika on a world map.

 Transforming Knowledge Sharing ◾ 97

Corpus_fact
Location
location_key user_key
longitude time_key
lattitude location_key
text
location_freq favorited
favorite_count
reply_to_ScrNm
status_source
retweet_count

Figure 3.15 The location dimension in the KBDW model.

the preventive and protective actions that should be taken into

serious consideration, as the country’s climate is suitable for
the thriving of the mosquitos that carry the virus.
The analysis explained above is a case-in-point of how the
analysis of one dimension of a corpus allows for converting
the individual tacit knowledge into the hidden collective tacit
knowledge using data analytics tools. The extraction of the col-
lective tacit knowledge permits us to obtain a deeper insight of
the data (Tweets collected in May and June 2017), as we are
able to explain the authors’ reasoning for content creation and
to better understand the dynamics of the Zika virus.
Figure 3.15 shows the logical design of a small segment
of a KBDW that applies to Vignette 3.4. For KM, it is more
useful to aggregate locations from the unique point level to
a larger geographical area. Areas can be city blocks, neigh-
borhoods, postal codes, counties, states, and so on. There is
then a unique location_key for each area, and all areas are
listed in the location table. Using a data warehouse with this
configuration, we could generate tables listing various statistics
for each location area unit and create, for instance, input data
for diffusion analysis, by running a straightforward SQL query
that draws information from the location table. An example of
such a location statistic is location_freq, or location frequency,
shown in Figure 3.15 as a derived attribute.

Topic Dimension
As community members communicate their ideas by exchanging oral or written
documents, word usage patterns tend to exhibit characteristics of a spontaneous
order: From long and seemingly flat lists of words, a structure of organized, socially
constructed, corpus-level topics emerges. The modeling of such topics has taken
98 ◾ Analytics and Knowledge Management

two main approaches: the generative, or probabilistic, approach, and the varia-
tional, or statistical, approach.
The generative approach assumes that document authors are already aware of a
number of topics that express issues relevant to their community, and are aware of
the probability distribution of each topic across all terms in the dictionary. They then
generate words for the documents they author by selecting a topic from a mix of top-
ics that characterizes the particular document, and selecting a term from the mix of
terms that characterizes the particular topic. Latent Dirichlet allocation (LDA), the
most widely-cited algorithm that follows the probabilistic approach to topic modeling,
uses a Bayesian approach with Markov chain Monte Carlo simulations to estimate the
parameters of the multinomial distributions of topics across the documents and terms
across the topics (Blei, Ng, & Jordan, 2003; Blei, 2012). The variational approach
assumes that humans acquire meaning by being repeatedly exposed to multiple exam-
ples of documents containing such meaning (Kintsch & Mangalath, 2011). Because
of a need to minimize cognitive effort, they create a mental space that quantifies the
patterns of co-occurrence of terms within similar contexts as linear combinations,
or vectors, of terms and documents. Terms and documents are then projected into
that space in a lower dimensionality that drops thousands of dimensions that describe
the specific document content and keeps only a few abstract concepts that describe
terms and documents in a broad sense. Latent Semantic Analysis (LSA), the most
widely-cited algorithm that follows the variational approach to topic modeling, uses
the matrix operation of singular value decomposition to project the original term fre-
quency matrix to a space of principal components that explain maximum variation
using a minimum number of dimensions (Deerwester et al., 1990; Dumais, 2004).
The two approaches seem to be somewhat complementary, since LDA explains how
documents are produced from topics without explaining how topics were acquired and
LSA explains how topics are acquired without explaining how they were produced.
And, in practice, the two algorithms often extract very similar topics after process-
ing the same corpus. We now continue with Vignette 3.5, where we follow the LSA
approach and perform text analytics to extract topics from a corpus of Tweets.

VIGNETTE 3.5 Topics in Tweets on

International Women’s Day 2017
We continue the scenario in Vignette 3.3, addressing Tweets
about IWD, which is held globally on March 8 of each year
as a celebration of the political and social achievements of
women. After compiling the set of 82,857 IWD Tweets as
described in Vignette 3.3, we followed established text ana-
lytic preparation and term filtering steps (see, e.g., Coussement
and Van Den Poel, 2008). Since our goal was to understand
the latent semantic structure of our corpus, we implemented
 Transforming Knowledge Sharing ◾ 99

the factor analysis variant of LSA (Sidorova, Evangelopoulos,

Valacich, & Ramakrishnan, 2008; Evangelopoulos, Zhang, &
Prybutok, 2012). In our analysis, we followed the main LSA
steps outlined in Kulkarni, Apte, and Evangelopoulos (2014), as
well as Evangelopoulos, Ashton, Winson-Geideman, & Roulac
(2015) as listed below.

◾ LSA step 1, term frequency matrix: The first step of the

analysis is to compile a term frequency matrix, also
known as the vector space model, by using software that
parses each passage into individual words and removes
trivial English terms such as “the,” “of,” “and,” and so on,
as well as low-frequency terms. For the implementation
of this step, we used the commercial software package
SAS® Text Miner, available free of charge to academics
(SAS, 2017). Following standard text mining practice, we
weighted the raw term frequencies using a transformation
called inverse document frequency, which promotes the
frequency of rare terms and discounts the frequency of
common terms and stemmed the terms to combine terms
such as organize, organized, organizes, organization, and
so on. We also performed n-gram analysis to identify
groups of words that tend to appear together as a phrase
and use them as a single, composite term.
◾ LSA step 2, decomposition: The term frequency matrix
was subjected to singular value decomposition (SVD). The
output of this step included a set of term loadings and a
set of document loadings on the extracted latent semantic
dimensions.
◾ LSA step 3, dimensionality selection and rotation of
the dimensions: Following Kulkarni et al. (2014) and
Evangelopoulos et al. (2015), we produced a scree plot of the
eigenvalues. This is shown in Figure 3.16. An “elbow” point
is identified at k = 4, which means four topics are retained.
◾ LSA step 4, factor labelling: To understand the extracted
topics, we examined the two sets of factor loadings, that
is, the loadings for terms and the loadings for Tweets (doc-
uments). Each topic was then labeled based on the related
high-loading terms and Tweets. Overall, the factors were
easy to understand, as the high-loading Tweets for each
topic were very similar to each other, and we were able to
complete the labeling process without controversy.

Topic analysis for the IWD 2017 Tweets produced four overarch-
ing topics. The first topic—Here is to Strong Women—included
100 ◾ Analytics and Knowledge Management

500

450

400

350

300
Eigenvalue

250

200

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Dimension

Figure 3.16 Scree plot of the first 20 eigenvalues for the IWD 2017 Tweets.

terms such as “strong,” “strong woman,” “know,” “beautiful,”

and “raise.” Many of the individual Tweets associated with this
topic were postings of the phrase “Here’s to strong women.
May we know them. May we be them. May we raise them.”
The second topic—Celebrate Women—comprised individual
Tweets that mostly related to the sentiment of IWD in gen-
eral and included terms such as “celebrate,” “world,” “today,”
“good,” and “amazing.” The third topic—Love—was generated
by various messages of support for women, which included
terms such as “love,” “world,” “amazing,” “life,” and “inspire.”
The final topic—Wear Red in Support—directly reflects the
popularity of the Day Without Women strike, as indicated by
terms such as “today,” “red,” “wear,” “support,” and “work.”
Table 3.3 is a list of the topics, terms, and the number of docu-
ments associated with each topic.
Figure 3.17 shows the logical design of a small segment of
a KBDW that applies to Vignette 3.5. Since the topics were
extracted from the corpus through text analytics, table topic is
modeled as a derived dimension. Topic attributes include its ID
field topic_key, the label, a list of top-loading terms, and the total
count of all documents in the corpus that are associated with
the topic. If a data warehouse with this configuration is built,
we can generate Table 3.3 by running a straightforward SQL
query that simply lists the contents of the topic table. Regarding
 Transforming Knowledge Sharing ◾ 101

Table 3.3 Topics and Terms for IWD 2017 Tweets and Their Associated
Document Counts
Doc.
Topic Label Top Loading Terms Count

T1 Here is to +strong,+strong woman,+know, 4157

strong women beautiful,+raise

T2 Celebrate +celebrate,+world, today,+good, 5641

women amazing

T3 Love +love,+world, amazing,+life,+inspire 3650

T4 Wear red in today,+red,+wear,+support,+work 5065

support

Topic
Corpus_fact
topic_key
author_key label
time_key top_terms
location_key doc_count
tweet_id
text
favorited Tweet_by_topic
favorite_count
topic_key
reply_to_SN
tweet_id
status_source
topic_strength
retweet_count

Figure 3.17 Topic as a derived dimension in the KBDW model.

the association between Tweets and topics, an initial modeling

approach might record topic_key and a corresponding topic_
strength as derived attributes in the fact table. This, however,
assumes that each Tweet can only be associated with one topic.
In reality, Tweets may cross-load on multiple topics, which
highlights one of the strengths of topic modeling compared to,
say, clustering, where each document is typically assigned to a
unique cluster. Therefore, Figure 3.17 shows a Tweet_By_Topic
table that models the many-to-many relationship between the
Corpus_Fact and Topic tables. Topic strength can be based on a
factor loading scale, which shows the degree to which a Tweet is
correlated with a topic, or derived in the form of a binary indica-
tor that tags the Tweet as “relevant” to the specific topic.
102 ◾ Analytics and Knowledge Management

Topic-Time Interaction
Topics do not have a static preference among TBC members. In a constantly
changing environment, the community members adapt discourse and let the
topics in the corpus wax and wane. In KBDW terms, the dynamic behavior
of topics across time can be studied by considering the interaction between
the topic and time dimensions. Vignette 3.6 provides an example of this
interaction.

VIGNETTE 3.6 International Women’s

Day 2017: Tweet Topics by Hour
We continue the scenario in Vignettes 3.3 and 3.5, address-
ing Tweets about IWD 2017. After compiling the set of 82,857
IWD Tweets as described in Vignette 3.3, we performed
topic extraction as described in Vignette 3.5. At this point,
each Tweet was associated with one of the 24 hours of the
day and one or more of the four extracted topics. The next
step was to perform spreadsheet manipulations of the table of
82,857 Tweets that included an hour column and the 82,857-
by-4 matrix that provided the Tweet-to-topic association
and produced a 24-by-4 cross-tabulation of Tweet counts
per hour, by topic, shown in Table 3.4. Tweets for hours 2,
4, 6, 8, and 10 were not collected, therefore count data for
these hours was interpolated and then rounded to the nearest
integer. Such a cross-tabulation (also known as contingency
table) can help us identify certain elements of the topics’
dynamic behavior by performing additional analytics. Since
the columns of the table are essentially distributions of each
topic across the 24 hours of the day, one analytic option is
to perform chi-square tests that compare these distributions.
The test is highly significant, producing a Pearson chi-square
statistic equal to 259.832, and a p-value less than 0.001. This
indicates that the distributions of the four topics across the
24 hours, shown in Figure 3.18 as trend lines, are significantly
different.
The four trend lines shown in Figure 3.18 help us tell
the story of IWD 2017: In the late morning hours, calls
 Transforming Knowledge Sharing ◾ 103

Table 3.4 Cross-Tabulation of the Four Topics by Hour of the Day

Hour T1 T2 T3 T4

0 12 25 7 17

1 10 12 10 14

2 19 48 17 32

3 27 84 23 49

4 40 95 37 57

5 53 105 50 65

6 84 153 81 96

7 114 201 112 127

8 170 331 159 210

9 225 461 205 293

10 285 501 245 329

11 345 541 285 365

12 300 416 276 399

13 283 328 234 392

14 554 756 461 762

15 278 379 257 378

16 470 600 387 534

17 222 296 220 299

18 55 61 46 58

19 218 291 191 271

20 238 281 228 244

21 255 243 195 245

22 233 277 225 263

23 265 284 238 290

104 ◾ Analytics and Knowledge Management

900
800
700
600
Tweet count

500
400
300
200
100
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of the day
Here is to strong women Celebrate women
Love Wear red in support

Figure 3.18 Time trends for the four IWD Tweet topics over the 24-hour day on
March 8, 2017.

to celebrate women dominate the discourse, but, as we

move into the early afternoon and various demonstration
and celebration events unfold, calls to wear red in support
take over. We expect, of course, these patterns to be com-
mon knowledge among the most organized community
members who participated—and Tweeted—in IWD 2017,
but that is community tacit knowledge. With the help of
SMAs, we can extract this knowledge from the community
corpus and make it explicit in the form of Table 3.4 and
Figure 3.18.
Going back to the design of our KBDW, we note that the
cross-tabulation shown in Table 3.4 can be generated in a
more straightforward way if, instead of spreadsheet manipula-
tions, we used database operations. The hour attribute of the
time dimension shown in Figure 3.13 associates each Tweet
with one of the 24 hours of the day, and the Tweet_By_Topic
table shown in Figure 3.17 associates each Tweet with one
or more of the four extracted topics. Figure 3.19 combines
these segments to model the indirect association between
the time and topic dimensions through the fact and tweet-
by-topic tables. Table 3.4 can now be generated by running a
SQL query on these tables.
 Transforming Knowledge Sharing ◾ 105

Corpus_fact
Topic
Time author_key
time_key topic_key
time_key label
date location_key
tweet_id Tweet_by_topic top_terms
month doc_count
date text topic_key
year favorited tweet_id
hour favorite_count topic_strength
reply_to_SN
hourly_freq status_source
retweet_count

Figure 3.19 The dimensions of topic and time, and their relationship in the
KBDW model.

Opinion Dimension
Textual data can generally be divided into the categories of fact or opinion, with
facts defined as “objective expressions about entities, events, and their properties,”
and opinions defined as “subjective expressions that describe people’s sentiments,
appraisals, or feelings toward entities, events, and their properties” (Liu, 2010,
p. 627). Sentiments, therefore, represent a specific type of textual opinion. The study
of opinions is important, because individuals and organizations seek opinions (or
word-of-mouth recommendations) when engaged in decision-making, and because
an individual’s opinion reflects the knowledge that he or she has gleaned from vari-
ous experiences. The process of uncovering the sentiments conveyed by documents is
called sentiment analysis or opinion mining, which is defined as “the computational
study of opinions, sentiments, and emotions expressed in text” (Liu, 2010, p. 629).
Opinionated texts contain an expression on their left and either an explicit
(overt) or implicit (implied) opinion on their left (Liu, 2010, p. 649). Opinions can
be expressed on a gamut of things, including people, places, organizations, events,
and topics. In sentiment analysis, the entity targeted by an opinion is termed its
object (Liu, 2010, p. 629). Objects are comprised of components (parts) and attri-
butes (properties), which can be further deconstructed into continuous layers of
subcomponents and subattributes, such that the objects of opinionated texts can
be depicted using “part of” the relationships represented by taxonomies, hierar-
chies, and trees. In addition to the object of the opinion, sentiment analysis seeks to
uncover its orientation (i.e., whether it is positive, negative, or neutral) and its source
(i.e., the opinion holder, or the person or organization that originally expresses the
opinion). Analysts seeking to conduct sentiment analysis can either find public opin-
ions about a specific object or find opinions from specific sources about an object.
The advent of the internet has been paramount in enabling the study of opinions,
and the popularity of user-generated content that began with Web 2.0 has made the
mass opinions conveyed via product reviews, forums, discussion groups, and blogs
106 ◾ Analytics and Knowledge Management

readily available for analysis (Liu, 2010, p. 627). Though an opinion holder’s sentiment
might be explicit in a product review, it is likely more implicit in longer texts and more
conversational documents, such as Twitter posts. Therefore, specialized techniques and
applications are required to uncover the underlying sentiments of these documents.
The feature-based sentiment analysis model (Hu & Liu, 2004; Liu, 2006; Liu,
Hu, and Cheng, 2005) instructs researchers to mine direct opinions for discover-
ing all quintuple groupings of opinion objects, object features, polarity, holders,
and time and identifying synonyms for words and phrases that convey opinions
that occur within the documents (Liu, 2010, p. 632). Results can then be conveyed
using opinion summaries and visualized with bar graphs and pie charts (Pang &
Lee, 2004). However, to successfully implement the model, researchers must also
engage in sentiment classification and the generation of an opinion lexicon.
Sentiment classification seeks to classify opinionated documents as expressing
positive, negative, or neutral opinions. The sentiment classification process is five-
fold and consists of identifying terms and their frequencies and then tagging those
terms according to their parts of speech. From there, opinion words and phrases are
isolated and evaluated for both their syntactic dependency (the degree to which the
interpretation of a word or phrase is dependent on its surrounding words or phrases)
and the effects of any negation in the phrase (the degree to which the inclusion of
a negative word, such as “not” or “but,” affects whether the expressed opinion is
positive or negative) (Liu, 2010, p. 638). For example, application rules dictate the
following:

◾ Negative adjective = negative opinion

◾ Positive adjective = positive opinion
◾ Negation word + negative adjective = positive opinion
◾ Negation word + positive adjective = negative opinion

An opinion lexicon is the equivalent to a dictionary to which the text of the opin-
ionated documents will be compared, and its generation depends on the identifica-
tion and compilation of opinion phrases and idioms, in addition to base-type and
comparative-type positive and negative opinion words (Liu, 2010, pp. 641–642).
Positive-opinion words express desired states, while negative-opinion words express
undesired states. Base-type opinion words are basic adjectives, such as “good” or “bad,”
while comparative-type opinion words express opinions of comparison or superlatives
using terms such as “better,” “worse,” “best,” and “worst,” which are derived from their
associated base-type words. Comparative-type opinion words are unique because they
state an opinion that two objects differ from each other on a certain feature, rather
than stating a simple opinion on a single object. Documents can contain four types of
comparative relationships, which can be subcategorized into gradable and nongradable
comparisons. Gradable comparisons express opinions of non-equality (that the quali-
ties of one object are greater or less than those of another), equality (that the qualities of
two objects are the same or similar), or superlative (that one object is the best or worst
 Transforming Knowledge Sharing ◾ 107

of all comparable objects in respect to a quality). Nongradable comparisons convey

difference, but without any positive or negative connotations.
To collect and compile the opinion words and phrases for the lexicon, analysts
may employ either a dictionary-based or corpus-based approach (Liu, 2010, p. 642).
When utilizing a dictionary-based approach, researchers manually collect small sets
of opinion words and then search for their synonyms and antonyms via a platform
such as WordNet (Hu & Liu, 2004; Kim & Hovy, 2004). The ongoing accumula-
tion of synonyms and antonyms expands the initial set of words and phrases, and
this bootstrapping-type process continues until no new words are found. Finally, the
analysts review the complete set of opinion words and phrases and manually remove
or correct any errors. The implementation of the corpus-based approach begins with
a list of seed opinion adjectives, which are used, along with sets of linguistic con-
straints and conventions regarding connectives and sentence structures, to identify
additional opinion words and their orientations. The reliance upon rules of syntax
means that conjunctions are especially important for establishing sentiment consis-
tency via the corpus-based approach. For example, Tweets containing “and” indicate
complimentary opinions between the phrases that the conjunction connects. On
the other hand, phrases connected by words such as “or” or “but” might not convey
similar opinions. Once words and phrases are collected and compiled and sentiment
consistency is established, the final set is clustered on a graph to produce positive
and negative subsets for the lexicon.
Opinions are expressions, which means they are necessarily conceptual; there-
fore, they represent concepts that opinionated text sources can express in a myriad
of ways (Liu, 2010, p. 649). Thus, regardless of the lexicon-generation method
employed, sentiment analysts must be aware of various linguistic rules that gov-
ern the structuring of sentences, if they are to accurately interpret the underlying
sentiment that the opinionated text is expressing. For example, instead of purely
negative or positive opinions, text documents might refer to deviations from an
expected norm. In such instances, documents referring to an outcome that is within
the range of desired values can be interpreted as positive expressions, whereas docu-
ments referring to outcomes outside of the desired range of values (either too low
or too high) can be interpreted as negative expressions. Opinionated texts might
also refer to increases and decreases in quantities of opinionated items, such that:

◾ Decrease in a negative item = Positive opinion

◾ Decrease in a positive item = Negative opinion
◾ Increase in a negative item = Negative opinion
◾ Increase in a positive item = Positive opinion

Though this chapter deals specifically with expressions of positive, negative, and
neutral sentiment, opinion analyses, in general, can be undertaken to assess more
complex phenomena, including emotions (Aman & Szpakowicz, 2007) such
as joy, sadness, anger, fear, surprise, and disgust (Strapparava & Mihalcea, 2008;
108 ◾ Analytics and Knowledge Management

Chaffar & Inkpen, 2011), political leanings (Grimmer & Stewart, 2013; Ceron
et al., 2014), and media slant (Gentzkow & Shapiro, 2010).
Analyses that extend beyond opinion polarity must employ the tactics men-
tioned here, in addition to those for identifying, classifying, and cataloging a lex-
icon of words that align with the various emotions the analysts aim to uncover
(Strapparava & Mihalcea, 2008). This process requires that researchers uncover not
only the words that refer directly to emotional states (e.g., “fearful” or “excited”), but
also the ones that denote an indirect contextual reference (e.g., words like “laugh” or
“cry,” which indicate emotional responses to a stimulus). Then, the researchers pri-
marily employ the same methods for analysis. Regarding assessing political leanings
(e.g., Ceron et al., 2014), there is no doubt that parties, candidates, and news media
worldwide are interested in the promise of utilizing the opinions expressed on SNS
to predict the outcomes of elections by forecasting voting behavior. Advances in this
area even have the potential to render obsolete more traditional means of forecast-
ing, such as public opinion polling. Finally, the modern proliferation of perceived
media biases (Gentzkow & Shapiro, 2010) offers applications of opinion-analysis
techniques to discern whether text-based reports actually contain a media slant, or
whether such leanings are mere matters of readers’ and pundits’ interpretations.
Vignette 3.7 provides an example that illustrates a simple approach to opinion
mining, dictionary-based, sentiment analysis.

VIGNETTE 3.7 Seattle Tweets in

February 2017: Tweet Sentiment
As the residents of a certain city undertake their daily lives,
social media provides outlets to express their opinions and
feelings. Examining individual Tweets, positive or negative
sentiment is certainly influenced by the context of the discus-
sion. However, when Tweet sentiment is considered at a high
level of aggregation, individual reasons for positive or negative
feelings disappear, and what is left is the mood of an entire
community, affected by social events, local problems, or local
reasons to celebrate. To investigate the sentiment of Seattle
residents, we used the Twitter API and the R library twitteR
(Gentry, 2016) to collect all Tweets that mentioned Seattle over
a period of 20 days in February 2017. A total of 113.9 K Tweets
posted between 6 p.m. and midnight were collected. Of these,
6,754 Tweets, or about 6%, contained geo-location data.
We scored these Tweets for sentiment using the dictionary-
based sentiment analysis approach (Hu and Liu, 2004; Liu
2010). Examples of positive dictionary entries include words
such as accomplishment, amazing, approve, awesome, bliss,
 Transforming Knowledge Sharing ◾ 109

charming, comfort, and so on. Examples of negative diction-

ary entries include words such as alienate, allergy, annoying,
angry, badly, boring, careless, and so on. The 2,006 positive
entries and 4,784 negative entries compiled by Hu and Liu
(2004) were used for this purpose, and the analysis was done
in R using the Breen sentiment scoring algorithm. The results
of sentiment scoring are shown in Table 3.5 and Figure 3.20.
The sentiment scores are primarily positive.

Table 3.5 Frequencies of Sentiment

Scores in the Geolocated Sample
Sentiment Score Frequency

−4 0

−3 2

−2 25

−1 265

0 3156

1 2664

2 186

3 15

4 2

3500

3000

2500
Frequency

2000

1500

1000

500

0
−4 −3 −2 −1 0 1 2 3 4
Sentiment score

Figure 3.20 Histogram of sentiment scores in Seattle Tweets, February 2017.

110 ◾ Analytics and Knowledge Management

Corpus_fact
author_key
time_key
location_key
text
favorited
favorite_count
reply_to_SN Opinion
status_source opinion_key
retweet_count opinion_type
opinion_key
opinion_strength

Figure 3.21 Opinion as a derived dimension in the KBDW model.

Figure 3.21 shows the logical design of a small segment of a

KBDW that pertains to Vignette 3.7. Since the sentiments were
extracted from the corpus through text analytics, table opinion
is modeled as a derived dimension. Opinion attributes include
an ID field, opinion_key, and opinion_type, which could take
the value sentiment, as in Vignette 3.7, or other values, such as
joy or fear, that may apply to other types of opinion analysis.
The derived attribute opinion_strength, added to the fact table,
reflects the degree of positive or negative sentiment, the degree
of joy, and so forth. Assuming that a data warehouse with this
configuration is built, we can generate the frequency distribu-
tion table that produced the histogram shown in Figure 3.20 by
running a very straightforward SQL query that combines infor-
mation from the contents of the fact and the opinion tables.

Opinion-Location Interaction
Since human communities tend to self-organize and cluster together in space with
similar others, opinions can vary significantly across geographical locations. In
KBDW terms, the spatial dynamics of opinion can be studied by considering the
interaction between the opinion and location dimensions. Vignette 3.8 provides an
example of such interaction.

VIGNETTE 3.8 Seattle Tweets in February 2017:

Tweet Sentiment by Geographical Sector
How does a city resident’s mood vary by location? Following
up on the data collection of 113.9 K Tweets that mentioned the
 Transforming Knowledge Sharing ◾ 111

term, Seattle in a 20-day period in February 2017, described in

Vignette 3.7, we identified 6,754 Tweets (about 6% of all the
collected Tweets) that contained geo-location data. Since each
Tweet is already scored for sentiment as described in Vignette 3.7,
a spreadsheet that listed the 6,754 Tweets together with their
sentiment score and their geo-location codes was assembled.
We then proceeded with spatial analytics. After some initial
exploration of the longitude and latitude coordinates of the
6,754 Tweets, the boundaries of a rectangle that contains the
city of Seattle were defined by the four points of NW (‒122.406,
47.735), NE (‒122.245, 47.735), SE (‒122.245, 47.5345), and
SW (‒122.406, 47.5345), where NW stands for northwest and
so forth. This rectangular area was subsequently divided into
80 sectors, based on a 10-by-8 grid. With the boundaries of
each one of the 80 sectors defined, each geo-located Tweet was
classified under one of the 80 sectors x1y1, x1y2, and so forth.
The sector of this grid was used as the aggregation grain, and
summary statistics were obtained for each sector. Table 3.6 lists
the average sentiment score by sector, with the rows of the table
being laid out from north to south and the columns from west
to east so that the table places the sectors as they would appear

Table 3.6 Average Sentiment Score in the 80 Sectors in the 10-by-8 Grid of
Seattle Locations
West East

X1 X2 X3 X4 X5 X6 X7 X8

North Y10 0.48 0.11 0.50 0.43

Y09 0.25 0.78 0.55 0.63 1.00

Y08 ‒0.25 0.38 0.21 0.00 0.25 0.38 0.00 0.00

Y07 ‒0.10 0.17 0.00 0.35 0.22 0.38 0.43

Y06 0.06 0.44 0.30 0.50 0.50 0.82 0.25

Y05 0.29 0.29 0.22 0.37 0.50 0.40

Y04 0.40 0.50 0.27 0.50 0.50

Y03 1.17 0.36 0.38 0.32 0.49 0.00

Y02 0.23 0.07 0.00 ‒0.13 ‒0.08 0.22 0.43 2.00

South Y01 0.00 0.00 0.44 0.05 0.71 0.60 0.00

112 ◾ Analytics and Knowledge Management

0.0 0.2 0.4 0.6 0.8 1.0

0.0
0.2
0.4
0.6
0.8
1.0

Figure 3.22 Distribution of average sentiment scores across the grid of 80 Seattle
sectors.

on a map. Figure 3.22 was produced by overlaying an outline of

the Seattle map on a heatmap produced in R by first reading the
data in Table 3.6 as a matrix, and then executing the R function
image (sentiment). Dark gray sectors indicate smaller sentiment
scores and light gray sectors indicate larger (i.e., more positive)
sentiment scores. White sectors indicate no sentiment score data.
Going back to the design of our KBDW, we note that
Table 3.6 can be generated in a more straightforward way if,
instead of spreadsheet manipulations, we used database opera-
tions. In Figure 3.23 we show a segment of a KMDW model
that pertains to Vignette 3.8. The location grain is set to the sec-
tor level, and, for each sector, attributes longitude_bound1, lon-
gitude_bound2, latitude_bound1, and latitude_bound2 define
the bounds of the sector. Considering the average sentiment
of each location sector as a slowly changing attribute, we can
accommodate it as the derived attribute location_ave_sentiment
 Transforming Knowledge Sharing ◾ 113

Corpus_fact

author_key
Location time_key
location_key
location_key text
longitude_bound1 favorited
longitude_bound2 favorite_count
latitude_bound1 reply_to_SN Opinion
latitude_bound2 status_source
opinion_key
location_freq retweet_count
opinion_type
location_ave_sentiment opinion_key
opinion_strength

Figure 3.23 The dimensions of opinion and location, and their relationship in
the KBDW model.

within the location dimension. Values for this attribute would

have to be computed by executing a database query that com-
bines information from the opinion, location, and fact tables.
This is currently shown in Figure 3.23. However, if the local
sentiment is to be assessed periodically, then an additional
Location_By_Time table should be added, accommodating the
many-to-many relationship between location and time, and
recoding an opinion statistic (such the average sentiment of our
Vignette 3.8) for each location-sector-time-period combination.

Conclusion and the Road Ahead

We have introduced a dimensional model approach to the design of a knowledge
base for TBCs. From the theoretical point of view, we argue that a community of
humans who actively communicate with each other through social media postings
builds tacit knowledge that is embedded in a representative social media corpus.
We have then illustrated how different kinds of SMAs, including social network
analytics, time series analytics, text analytics, and spatial analytics, operate on
the corpus and convert its embedded tacit knowledge into explicit, in the form
of reports, tables, and visualizations. We have finally proposed a data warehouse
model for a knowledge base, which starts with core data warehouse fact and dimen-
sion tables that correspond to explicit knowledge, and iteratively builds additional
dimension and fact tables that uncover the community’s tacit knowledge and
make it accessible to the knowledge base user. With the proposed KBDW model,
tacit knowledge is now made explicit, as it is stored, managed, and easily shared.
Figure 3.24 presents a schematic representation of this proposed framework, where
a CoP builds socially constructed knowledge and embeds it in a corpus. With the
114 ◾ Analytics and Knowledge Management

Tacit collective knowledge Explicit knowledge

Community
of practice

NoSQL
repository Reports
(Corpus) SQL
queries

Data warehouse Visualization

(KBDW) algorithms
Analytics,
SQL queries Visualizations
and ETL

Figure 3.24 Framework for a corpus-based knowledge management system and

transition from tacit collective knowledge to explicit knowledge.

help of analytics and a data warehouse architecture to accommodate the knowledge

base, a corpus-based KM system can systematically transform tacit knowledge to
explicit. This process is presented in Figure 3.24 as a transition from left to right.
Without the proposed KBDW dimensional model, all the analytics we have
illustrated in our vignettes would produce separate reports that would include our
Tables 3.1 through 3.6, stored as separate spreadsheets, and the corresponding visu-
alizations. However, this approach would keep us in the early days of information
and KM, when knowledge was stored in the form of disconnected spreadsheets and
unstructured reports. A modern KM system needs a well-designed knowledge base
that can be queried to produce reports on the fly. We believe that a dimensional
model, such as the one we have presented in this chapter, can serve this purpose well.
The partial designs of a KBDW shown in Figures 3.8, 3.11, 3.13, 3.15, 3.17,
3.19, 3.21, and 3.23, can be extended to accommodate a variety of settings. The
Twitter user can be replaced by the author of any document or the creator of any
artifact. The case of Tweets examined in this chapter did not require any dedicated
treatment, curation, or management of individual documents, and this is why we
modeled the Tweets as transactional entries in the fact table. In different settings,
such as rare document collections or art collections, an upgrade of the document
entity to its own dimension might be appropriate. New dimensions can be added
to account for document clusters, other types of document categories, or schools
 Transforming Knowledge Sharing ◾ 115

Corpus_fact
Author/
Creator Time
author_key
Author_by_ time_key
topic location_key
author_key Location document_key
topic_key platform_key Platform/
assoc_strength storage_ref Medium
Topic access_rights
presentation_details
Document_by content_details
_topic metadata Cluster/
Document/ Category
topic_key cluster_key
document_key Artifact
cluster_centrality
assoc_strength opinion_key School of
opinion_strength thought/
Opinion school_key Community
school_position
impact_strength

Figure 3.25 Extensions to the dimensional model for corpus-based knowledge

management.

of creative thought associated with the document or artifact. Additional fact tables
can account for interactions between the author and topic dimensions, or the
document and topic dimensions. Some of these ideas are depicted in Figure 3.25.
We conclude this chapter by reiterating the importance of communities of prac-
tice, the collective knowledge they accumulate, and the proper management of their
knowledge through well-designed KM systems. We believe that KM researchers and
practitioners need to pay more attention to community knowledge. To put it in the
words of Gruber, such collective knowledge should be “taken seriously as a scientific
and social goal” (Gruber, 2008, p. 5). We hope our chapter has shed some light on
analytic and modeling considerations as one begins to accomplish this goal.

Summary
In this chapter, we examine the knowledge that develops within a community of
Twitter users. Focusing our discussion on a community of users that is built around
a certain social or scientific interest, and a subset of actively involved contributors, we
view such a Twitter-based community (TBC) as an online CoP. We view a corpus of
Tweets produced by the TBC as the community’s store of common knowledge. Using
various kinds of SMAs that operate on such a corpus, we uncover the collective tacit
knowledge that is embedded in it and discuss the process of its transfer to a data ware-
house. We present modeling elements of this data warehouse based on the dimensions
of user, time, location, topic, and opinion. We then discuss how physical database
designs would include these dimensions and their interactions as database tables and
116 ◾ Analytics and Knowledge Management

how the execution of simple database queries would then transform the TBC’s tacit
collective knowledge into an explicit form. We include eight illustrative vignettes that
examine various aspects of collective knowledge built within TBCs that discuss the
Zika virus, International Women’s Day, and the city of Seattle.

References
Adamson, C., & Venerable, M. (1998). Data warehouse design solutions. New York, NY:
John Wiley & Sons.
Aman, S., & Szpakowicz, S. (2007). Identifying expressions of emotion in text. In
V. Matoušek and P. Mautner (Eds.), Text, speech and dialogue, lecture notes on artificial
intelligence (Vol. 4629, pp. 196–205). Berlin, Germany: Springer-Verlag.
The Atlantic. (2017). Brazil declares an end to its Zika health emergency. Retrieved from
https://www.theatlantic.com/news/archive/2017/05/brazil-ends-zika-emergency/526509/.
BBC News. (2017). Zika virus: Brazil says emergency is over. BBC News, May 12, 2017.
Retrieved from http://www.bbc.com/news/world-latin-america-39892479.
Berger, P.L., & Luckmann, T. (1966). The social construction of reality. New York, NY:
Random House.
Blei, D.M. (2012). Probabilistic topic models. Communications of the ACM 55(4), 77–84.
Blei, D.M., Ng, A., & Jordan, M. (2003). Latent dirichlet allocation. Journal of Machine
Learning Research, 3, 993–1022.
Bodnar, T., & Salathé, M. (2013). Validating models for disease detection using twitter.
Proceedings of the 22nd International Conference on World Wide Web Rio de Janeiro,
Brazil (pp. 669–702). doi:10.1145/2487788.2488027.
Bowerman, B.L., O’Connell, R., & Koehler, A. (2005). Forecasting, time series, and regres-
sion (4th ed.). Stamford, CT: Thomson Learning.
Burns, T., & Stalker, G. M. (1961). The management of innovation. London, UK: Tavistock.
Carneiro, H.A., & Mylonakis, E. (2009). Google trends: A web-based tool for real-time
surveillance of disease outbreaks. Clinical Infectious Diseases, 49(10), 1557–1564.
CDC (2017). [Zika] Symptoms. Centers for disease control and prevention. Retrieved from
https://www.cdc.gov/zika/symptoms/symptoms.html.
Ceron, A., Curini, L., Iacus, S.M., & Porro, G. (2014). Every Tweet counts? How sentiment
analysis of social media can improve our knowledge of citizens’ political preferences
with an application to Italy and France. New Media & Society, 16(2), 340–358.
Chaffar, S., & Inkpen, D. (2011). Using a heterogeneous dataset for emotion analysis in
text. In C. Butz & P. Lingras (Eds.), Advances in artificial intelligence, lecture notes in
artificial intelligence (Vol. 6657, pp. 62–67). Berlin, Germany: Springer-Verlag.
Chenoweth, E., & Pressman, J. (2017, February 7). Analysis: This is what we learned
by counting the women’s marches. Washington Post. Retrieved from https://www.
washingtonpost.com/news/monkey-cage/wp/2017/02/07/this-is-what-we-learned-by-
counting-the-womens-marches/.
Coussement, K., & Van Den Poel, D. (2008). Improving customer complaint management
by automatic email classification using linguistic style features as predictors. Decision
Support Systems, 44(4), 870–882.
Culotta, A. (2010). Towards detecting influenza epidemics by analyzing Twitter
messages. Proceedings of the First Workshop on Social Media Analytics, USA.
doi:10.1145/1964858.1964874.
 Transforming Knowledge Sharing ◾ 117

Czarniawska-Jerges, B. (1992). Exploring complex organizations: A cultural perspective.

Newbury Park, CA: SAGE Publications.
Davenport, E. & Hall, H. (2002). Organizational knowledge and communities of practice.
Annual Review of Information Science and Technology, 36(1), 170–227.
Davenport, E. (2001). Knowledge management issues for online organizations: Communities
of practice as an exploratory framework. Journal of Documentation, 57(1), 61–57.
De Long, D.W., & Fahey, L. (2000). Diagnosing cultural barriers to knowledge man-
agement. Academy of Management Executive, 14(4), 113–127. doi:10.5465/
AME.2000.3979820.
Deerwester, S., Dumais, S., Furnas, G., Landauer, T., & Harshmans, R. (1990). Indexing by
latent semantic analysis. Journal of the American Society for Information Science, 41(6),
391–407.
Diederich, J., Ruhmann, I., & May, M. (1987). KRITON: A knowledge-acquisition tool
for expert systems. International Journal of Man-Machine Studies, 26, 29–40.
Duffy, M.R., Chen, T.H., Hancock, W.T., & Hayes, E.B (2009). Zika virus outbreak on
Yap Island, Federated States of Micronesia. The New England Journal of Medicine,
360, 2536–2543. doi:10.1056/NEJMoa0805715.
Dumais, S.T. (2004). Latent semantic analysis. Annual Review of Information Science and
Technology, 38, 189–230.
Evangelopoulos, N., Ashton, T., Winson-Geideman, K., & Roulac, S. (2015). Latent
semantic analysis and real estate research: Methods and applications. Journal of Real
Estate Literature, 23(2), 355–380.
Evangelopoulos, N., Magro, M., & Sidorova, A. (2012). The dual micro/macro inform-
ing role of social network sites: Can twitter macro messages help predict stock
prices? Informing Science: The International Journal of an Emerging Transdiscipline,
15, 247–268. Retrieved from http://www.inform.nu/Articles/Vol15/ISJv15p247-
268Evangelopoulos0630.pdf.
Evangelopoulos, N., Zhang, X., & Prybutok, V.R. (2012). Latent semantic analysis: Five meth-
odological recommendations. European Journal of Information Systems, 21(1), 70–86.
Freeman, L.C. (1978). Centrality in social networks conceptual clarification. Social
Networks, 1, 215–239.
Gentry, J. (2016, August 29). Package ‘TwitteR’. CRAN repository. Retrieved from https://
cran.r-project.org/web/packages/twitteR/twitteR.pdf, accessed on July 3, 2017.
Gentzkow, M., & Shapiro, J.M. (2010). What derives media slant? Evidence from U.S.
daily newspapers. Econometrica, 78(1), 35–71.
Girvan, M., & Newman, M. E. J. (2002). Community structure in social and biologi-
cal networks. Proceedings of the National Academy of Sciences of the United States of
America 99(12), 7821–7826. doi:10.1073/pnas.12265379.
Gold, A.H., Malhotra, A., & Segars, A.H. (2001). Knowledge management: An organizational
capabilities perspective. Journal of Management Information Systems, 18(1), 185–214.
Gonzalez, A.J., & Dankel, D.D. (1993). The engineering of knowledge-based systems: Theory
and practice. Englewood Cliffs, NJ: Prentice Hall.
Granovetter, M.S. (1973). The strength of weak ties. American Journal of Sociology, 78(6),
1360–1380.
Granovetter, M.S. (1983). The strength of weak ties: A network theory revisited. Sociological
Theory, 1, 201–233.
Grimmer, J., & Stewart, B.M. (2013). Text as data: The promise and pitfalls of automatic
content analysis methods for political texts. Political Analysis, 21, 267–297.
118 ◾ Analytics and Knowledge Management

Gruber, T. (2008). Collective knowledge systems: Where the social web meets the Semantic
Web. Journal of Web Semantics, 6(1), 4–13.
Grudin, J. (1994). Computer supported cooperative work: History and focus. IEEE, 27 (5), 15–17.
Gunawardena, C.N., Hermans, M.B., Sanchez, D., Richmond, C., Bohley, M., & Tuttle, R.
(2009). A theoretical framework for building online communities of practice with
social networking tools. Educational Media International, 46(1), 3–16.
Hemsley, J., & Mason, R.M. (2013). Knowledge and knowledge management in the social media
age. Journal of Organizational Computing and Electronic Commerce, 23(1–2), 138–167.
Hu, M., & Liu, B. (2004). Mining and summarizing customer reviews. Proceedings of
the ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)
(pp. 168–177). Seattle, WA.
Ibarra, H., & Andrews, S. (1993). Power, social influence, and sense making: Effects of
network centrality and proximity on employee perceptions. Administrative Science
Quarterly 38(2), 277–303. doi:10.2307/2393414.
Iedema, R. (2003). Discourses of post-bureaucratic organization. Amsterdam, the Netherlands:
John Benjamins Publishing Company.
International Women’s Day. (2017). About International Women’s Day (March 8).
Retrieved from https://www.internationalwomensday.com/About
Jubert, A. (1999). Communities of practice. Knowledge Management, 3(2), 1999.
Kabir, N., & Carayannis, E. (2013). Big data, tacit knowledge and organizational competi-
tiveness. Journal of Intelligence Studies in Business, 3(3), 54–62.
Khan, G., F. (2015). Seven layers of social media analytics: Mining business insights from
social media text, actions, networks, hyperlinks, apps, search engine, and location data.
Lexington, KY: CreateSpace Independent Publishing Platform.
Kim, S.M., & Hovy, E. (2004). Determining the sentiment of opinions. Proceedings of the
International Conference on Computational Linguistics (COLING). Geneva, Switzerland.
Kimball, R., & Ross, M. (2002). The data warehouse toolkit, (2nd ed.). New York, NY: John
Wiley & Sons.
Kintsch, W., & Mangalath, P. (2011). The construction of meaning. Topics in Cognitive
Science, 3(2), 346–370.
Kulkarni, S., Apte, U., & Evangelopoulos, N. (2014). The use of latent semantic analysis in
operations management research. Decision Sciences, 45(5), 971–994.
Lancichinetti, A., & Fortunato, S. (2009). Benchmarks for testing community detection
algorithms on directed and weighted graphs with overlapping communities. Physical
Review E, 80, 1–9. doi:10.1103/PhysRevE.80.016118
Lancichinetti, A., Fortunato, S., & Radicchi, F. (2008). Benchmark graphs for test-
ing community detection algorithms. Physical Review E, 78, 1–6. doi:10.1103/
PhysRevE.78.046110
Larsson, R., Bengtsson, L., Henriksson, K., & Sparks, J. (1998). The interorganiza-
tional learning dilemma: Collective knowledge development in strategic alliances.
Organization Science, 9(3), 285–305.
Lave, J. (1988). Cognition in practice: Mind, mathematics, and culture in everyday life. New
York, NY: Cambridge University Press.
Lave, J. (1991). Situated learning in communities of practice. In L. Resnick, J. M. Levine, &
S. D. Teasley (Eds.), Perspectives on socially shared cognition (pp. 63–82). Washington,
DC: American Psychological Association.
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation.
Cambridge, MA: Cambridge University Press.
 Transforming Knowledge Sharing ◾ 119

Leonard, D., & Sensiper, S. (1998). The role of tacit knowledge in group innovation.
California Management Review, 40(3), 112–132. doi:10.2307/41165946.
Levin, D. Z., & Cross, R. (2004). The strength of weak ties you can trust: The mediating
role of trust in effective knowledge transfer. Management Science, 50(11), 1477–1490.
Liu, B. (2006). Web data mining: Exploring hyperlinks, contents, and usage Data. Berlin,
Germany: Springer-Verlag.
Liu, B. (2010). Sentiment analysis and subjectivity. In N. Indurkhya & F. J. Damereau (Eds.),
Handbook of natural language processing (pp. 627–666). Boca Raton, FL: CRC Press.
Liu, B., Hu, M., & Cheng, J. (2005). Opinion observer: Analyzing and comparing opin-
ions on the web. Proceedings of WWW, Chiba. Japan.
Mishra, N., Schreiber, R., Stanton, I., & Tarjan, R.E. (2007). Clustering social networks.
In A. Bonato & F.R.K. Chung (Eds.) Algorithms and models for the web-graph. WAW
2007. Lecture notes in computer science (vol. 4863). Heidelberg, Germany: Springer.
The New York Times. (2017, June 3). India acknowledges three cases of Zika Virus. The New
York Times. Retrieved from https://www.nytimes.com/2017/06/03/world/asia/india-
zika-virus.html.
Nonaka, I. (1991). The knowledge-creating company. Harvard Business Review 69(6), 96–104.
Nonaka, I., & Takeuchi, H. (1995). The knowledge-creating company: How Japanese compa-
nies create the dynamics of innovation. New York, NY: Oxford University Press.
Palmer, J. D., & Fields, A. N. (1994). Computer supported cooperative work. IEEE, 27(5), 15–17.
Panahi, S., Watson, J., & Partridge, H. (2012). Social media and tacit knowledge sharing:
Developing a conceptual model. World Academy of Science, Engineering and Technology
Index 64, International Journal of Social, Behavioral, Educational, Economic, Business
and Industrial Engineering 6(4), 648–655. http://scholar.waset.org/1307-6892/5672.
Pang, B., & Lee, L. (2004). A sentimental education: Sentiment analysis using subjectivity
summarization based on minimum cuts. Proceedings of the 42nd annual meeting on
Association for Computational Linguistics (p. 271). Barcelona, Spain: Association for
Computational Linguistics.
Pruitt, S. (2017, March 6). The surprising history of International Women’s
Day. The History Channel. Retrieved from http://www.history.com/news/
the-surprising-history-of-international-womens-day.
Radicchi, F., Castellano, C., Cecconi, F., Loreto, V., & Parisi, D. (2004). Defining and
identifying communities in networks. Proceedings of the National Academy of Sciences,
101, 2658–2663.
Rich, E., Knight, K., & Nair, S.B. (2009). Artificial intelligence, (3rd ed.). New Delhi,
India: Tata McGraw-Hill.
Riege, A. (2005). Three-dozen knowledge-sharing barriers managers must consider. Journal
of Knowledge Management, 9(3), 18–35.
Salathé, M., & Khandelwal, S. (2011). Assessing vaccination sentiments with online social
media: Implications for infectious disease dynamics and control. PLOS Computational
Biology, 7(10), e1002199.
Sanghani, R. (2017, March 8). What is international women’s day? The Telegraph. Retrieved
from http://www.telegraph.co.uk/women/life/international-womens-day-2017-did-start-
important/.
SAS (2017). SAS® OnDemand for Academics. Retrieved from http://support.sas.com/
software/products/ondemand-academics/.
Sharda, R., Delen, D., & Turban, E. (2018). Business intelligence, analytics, and data science:
A managerial perspective, (4th ed.). Upper Saddle River, NJ: Pearson Education.
120 ◾ Analytics and Knowledge Management

Sidorova, A., Evangelopoulos, N., Valacich, J.S., & Ramakrishnan, T. (2008). Uncovering
the intellectual core of the information systems discipline. MIS Quarterly, 32(3), 467–
482 & A1–A20.
Strapparava, C., & Mihalcea, R. (2008). Learning to identify emotions in text. Proceedings
of the 2008 ACM symposium on Applied computing, Fortaleza, Ceara, Brazil (pp.
1556–1560). New York, NY: ACM.
Sullivan, D. (2001). Document warehousing and text mining. New York, NY: John Wiley & Sons.
Tsui, E. (2003). Tracking the role and evolution of commercial knowledge management
software. In C.W. Holsapple (Ed.), Handbook of knowledge management. Heidelberg,
Germany: Springer-Verlag.
United Nations. (2017). International Women’s Day March 8. Retrieved from http://www.
un.org/en/events/womensday/.
University of Chicago. (2014). International Women’s Day History. Retrieved from https://
iwd.uchicago.edu/page/international-womens-day-history.
Walsh, J. P., & Ungson, G. R. (1991). Organizational memory. Academy of Management
Review, 16, 57–91.
Weick K. E. (1995). Sensemaking in organizations. Thousand Oaks, CA: SAGE Publications.
Wenger, E. (1998). Communities of practice: Learning, meaning, and identity. Cambridge,
MA: Cambridge University Press.
Wenger, E. (2000). Communities of practice and social learning systems. Organization, 7(2),
255–246.
Wenger, E., McDermott, R.A., & Snyder, W. (2002). Cultivating communities of practice: A
guide to managing knowledge. Boston, MA: Harvard Business Press.
Wenger, E.C., & Snyder, W.M. (2000). Communities of practice: The organizational fron-
tier. Harvard Business Review, 78(1), 139–145.
Wiig, K.M. (1997a). Knowledge management: An introduction and perspective. Journal of
Knowledge Management, 1(1), 6–14. doi:10.1108/13673279710800682.
Wiig, K.M. (1997b). Knowledge management: Where did it come from and where will it
go? Journal of Expert Systems with Applications, 13(1), 1–14.
World Health Organization. (2017). Zika virus infection—India. World health organization,
May 26, 2017. Retrieved from http://www.who.int/csr/don/26-may-2017-zika-ind/en/.
Xu, W.W., Sang, Y., Blasiola, S., & Park, H.W. (2014). Predicting opinion leaders in Twitter
activism networks: The case of the wisconsin recall election. American Behavioral
Scientist, 58(10), 1278–1293.
Yang, Z., Algesheimer, A., & Tessone, C.J. (2016). A comparative analysis of community detec-
tion algorithms on artificial networks. Scientific Reports, 6(30750). doi:10.1038/srep30750.
Zappavigna-Lee, M.S. (2006). Tacit knowledge in communities of practice. In Encyclopedia
of communities of practice in information and knowledge management (pp. 508–513).
Hershey, PA: IGI Global.
Zappavigna-Lee, M.S., & Patrick, J. (2004). Literacy, tacit knowledge, and organizational
learning. Proceeding of the 16th Euro-International Systematic Functional Linguistic
Workshop. Madrid, Spain.
Zappavigna-Lee, M.S., Patrick, J., Davis, J., & Stern, A. (2003). Assessing knowledge man-
agement through discourse analysis. Proceedings of the 7th Pacific Asia Conference on
Information Systems. Adelaide, South Australia.
Zarya, V. (2017). A brief but fascinating history of International Women’s Day. Fortune, March 7,
2017. Retrieved from http://fortune.com/2017/03/07/international-womens-day-
history.
Chapter 4

Data Analytics for

Deriving Knowledge
from User Feedback
Kuljit Kaur Chahal and Salil Vishnu Kapur

Contents
Introduction ......................................................................................................121
Collecting User Feedback ..................................................................................123
Analyzing User Feedback ...................................................................................124
Opinion Mining ...........................................................................................125
Link Analysis ................................................................................................127
User Feedback Analysis: The Existing Work.......................................................128
Deriving Knowledge from User Feedback..........................................................130
Data Management ........................................................................................133
Data Analytics ..............................................................................................134
Knowledge Management ..............................................................................136
Conclusions and Future Work ...........................................................................137
References .........................................................................................................138

Introduction
User feedback plays a critical role in the evaluation and enhancement of the qual-
ity of any organizational setup. User feedback is the meaningful information given
with the purpose of sharing experiences, suggesting improvements, and expressing
opinions regarding a system. In modern times, in the electronic online ecosystems,

121
122 ◾ Analytics and Knowledge Management

gathering user feedback is not a tedious process. However, there is a need to iden-
tify suitable methods to discover and interpret meaningful patterns and knowl-
edge from these huge datasets. Getting and then acting on user feedback is very
important for running an organization successfully in the present day competitive
market. In this chapter, we will identify the role of analytics in understanding user
feedback of an organizational setup.
The Merriam-Webster dictionary defines feedback as the transmission of evalua-
tive or corrective information about an action, event, or process to the original or con-
trolling source. User feedback is the reaction of a user after using a service or product.
The reaction can be positive, negative, or neutral depending upon the user’s experi-
ence with the service or product. Given the proliferation of the Internet and social
media, user feedback can be instantly captured. People now have unprecedented
opportunities to raise their opinions on public platforms. Public opinion is useful and
right for organizations as they can get to know about the success and failure of their
policies and products from the user feedback easily available on social media.
However, most of the time, people express their opinions indirectly. Using
natural languages, they mix facts with their opinions (Pawar et al. 2016). A fact is
an objective expression regarding an event, experience, policy, or product, whereas
an opinion is a subjective expression (of emotion, perceptions, or perspectives).
Analysis of this subjective expression of the opinion holder is a key consideration
while mining knowledge from publicly shared discourses. Opinion mining, or
sentiment analysis, is the discipline that combines the techniques from different
fields such as natural language processing, text mining, and information retrieval
to segregate facts from opinions as they appear in unstructured text. Extracting
user opinion from billions of user posts and then analyzing it to use it for decision-
making is an insurmountable task, which needs automatic, rather than manual,
ways to accomplish it successfully.
In the past, investigations reported in the research literature in this area have
been largely restricted to reviews of business establishments such as hotels and res-
taurants, or electronic products such as cameras, mobile phones, and so on (Hridoy
et al., 2015). Nowadays, there is tremendous interest in gauging public sentiment
during elections to understand their response toward electoral issues and interest in
different political parties and politicians (Mohammad et al., 2015). However, there
is a need to analyze public response regarding government policies or other issues of
public interest. Government policies sometimes draw a lot of flak from the general
public. Therefore, it will be interesting to watch the Twitter stream to see the public
opinion regarding a government policy. Data analytics of user feedback on social
media may indicate the public’s responsiveness as well as their interest in a govern-
ment’s policies. In this chapter, we analyze public sentiment as expressed on Twitter
regarding the Indian government demonetizing high currency notes in November,
2016 to curb corruption. We stress the importance of extracting knowledge from
user feedback, and then integrating and reusing this knowledge from the crowd in
the decision-making process.
 Data Analytics for Deriving Knowledge from User Feedback ◾ 123

The study looks at what is known and not known about user feedback and
seeks to identify the user feedback processes as a valuable resource for improved
understanding of user needs for better decision-making. This chapter suggests an
approach by which progress toward better integration of user feedback within the
decision-making process through a knowledge management system in an organiza-
tion can be realized. Such improvements will assist policy-makers because knowl-
edge obtained from user feedback can be preserved as an asset for future use with
the help of a knowledge management system.
The major goals of this chapter are as follows:

1. To propose a three-step approach for integrating user feedback into the

knowledge management system of an organization to support policy-makers
in better decision-making.
2. To demonstrate the proposed approach by applying it to the process of under-
standing public response to a change in a government policy. This example
is regarding a sudden change in currency (demonetization) by the Indian
government in November 2016.

The rest of the chapter is organized as follows: The next section discusses possible
ways to collect user feedback; the third section gives details of opinion mining and
data mining as techniques for user feedback analysis; the fourth section presents the
existing work for user feedback analysis; our proposed approach is mentioned in the
fifth section; and the last section concludes the chapter.

Collecting User Feedback

User feedback can be obtained through explicit or implicit means. In explicit
feedback, a user can provide ratings, opinions, or comments through question-
naires or interviews or post them on social media and blogs. Implicit feedback
involves observing user actions to understand the user’s interaction with a sys-
tem. We discuss the different ways of collecting user feedback in the following
paragraphs.

◾ Observing user actions: This is a kind of implicit feedback in which a user’s

feedback is inferred on the basis of his or her actions while interacting with a
system. An online shopping system may track the purchase history of a user.
Frequent buyers may be considered users with positive feedback. Another
example can be when a user visits a web page time and again, and spends
considerable time on the web page. The number of mouse clicks on a web
page, or a usage pattern of various buttons on a page can also indicate to some
extent the level of satisfaction a user may have regarding a system. However,
negative feedback is not truly reflected through implicit means. The system
124 ◾ Analytics and Knowledge Management

provider may misinterpret negative feedback or may simply ignore it if it does

not happen in the way he or she understands.
◾ Through questionnaires, surveys, interviews: A user is asked to fill a survey
form or answer a set of questions through an interview. Here the questions
are pre-prepared. Organizations often use online or paper-based surveys
to collect customer feedback. However, this fixed form approach is not an
efficient approach as people find it boring and, therefore give only half-
hearted responses. Moreover, only a very small number of customers can be
approached. These traditional data collection methods are time consuming as
well. Collecting data even of the order of thousands of responses is sometimes
impractical due to the surmounting effort involved in the process. Open
ended questions may receive weird answers. Moreover, results are often biased
as either only positive or negative responses are possible.
◾ User response in social media networks, forums, comments sections: Here
social media acts as an interface between users and governments or orga-
nizations. People post their opinions about products or services on social
networking sites such as Facebook or Twitter, and get a quick response
from the concerned product or service providers. For example, recently
a popular Indian television news personality Tweeted about her (bad)
experience with a mobile company, which was instantly followed by posts
from her followers with similar experiences. It was interesting to see an
immediate response from the mobile company. It instantly reacted to the
negative feedback and promptly promised to resolve the issues. Social
media facilitates companies to reach out to a large customer base instantly.
Opinion mining or sentiment analysis helps to reflect user response toward
a product or service. Previous research shows that data obtained through
social media is good indicator of public opinion, and reflects public senti-
ment accurately (Hridoy et al., 2015). However, with this we may miss the
responses of certain population segments, such as people who are not using
social media (like old people) people with limited literacy levels, or those
who are less tech-savvy.

Analyzing User Feedback

Social media is a public platform where users express their opinions regarding
products, services, or policies. Several online platforms such as social networks,
blogs, review sites, wikis, and so on fall in the ambit of social media. Due to the
arrival of Web 2.0 in 2000 and the widespread adoption of social media thereaf-
ter, users started generating huge volumes of content online. Businesses saw a lot
of potential in this content for targeted marketing policies. Social media captures
two types of data regarding its user base: user generated content and links or
interactions between the communicating entities. A social network is modeled
 Data Analytics for Deriving Knowledge from User Feedback ◾ 125

as a graph in which nodes represent the communicating entities (e.g., users), and
edges represent the relationships or links (interactions) between the communi-
cating entities. To understand user feedback, interested parties can analyze the
content part to get the users’ sentiments. Along with this, links help to gauge the
spread of user sentiment through the social interactions of the user. Therefore, it
is important to understand user sentiment through content analysis, augmented
with link analysis to get meaningful insights. This section discusses opinion min-
ing as an approach to content analysis followed by data mining to perform link
analysis.

Opinion Mining
Opinion mining, or sentiment analysis, is an emerging tactic to analyze user feed-
back (Appel et al. 2015). Sentiment analysis pertains to extracting user expression
from user feedback expressed in a natural language. It makes use of text analysis
and computational linguistics to identify and extract attitudes expressed in user
feedback. A user may give positive, negative, or neutral feedback. An opinion is
expressed by a person (opinion holder) who expresses a viewpoint (positive, nega-
tive, or neutral) about an entity (target object, e.g., person, item, organization,
event, policy, and service) (Khan et al., 2014). Therefore, an opinion broadly
has the following parts: (1) opinion holders: people who hold the opinions; (2)
sentiments: positive or negative; (3) opinion targets: entities and their features or
aspects. As opinion may change with time, Liu (2011) adds a dimension of time,
as well.
Therefore, an opinion must have five parts: Opinion holder, a target entity,
aspects or features of the target, sentiment, and time. Appel et al. (2015) give a for-
mal expression: an opinion is represented as a quintuple (ej, ajk, soijkl, hi, tl), where: ej
is a target entity, ajk is an aspect or feature of the entity ej; soijkl is the sentiment value
of the opinion from the opinion holder hi on feature ajk of entity ej at time tl; soijkl
is positive, negative, neutral, or can have more granular ratings; hi is an opinion
holder; and tl is the time when the opinion is expressed.
Identifying all parts of an opinion quintuple from a given user discourse is
a challenging problem. An opinion holder is a user who has expressed an opin-
ion. Sometimes the opinion holder is mentioned explicitly. Otherwise, it has
to be presumed to be of the author of the piece. A user may not express intent
explicitly and rather use pronouns and context to relate different parts of the
quintuple. Target entity can be a product or policy. Sentiment, also known as
polarity of opinion, tells the orientation of an opinion, such as positive, nega-
tive, or neutral. Furthermore, an opinion can be expressed in the form of a
document or a collection of sentences. It could also be that a single sentence
contains a multitude of opinions on different features of an object. There is a
need to develop aspect-based systems to identify sentiments in such situations
(Mohammad, 2015).
126 ◾ Analytics and Knowledge Management

Extracting an opinion from an unstructured text expressed in a natural lan-

guage involves some ground work. The following steps are involved in the process
of opinion mining (Appel et al., 2015):

Subjectivity classification: In the first step, a user’s statements are categorized into
opinionated and nonopinionated ones. An opinionated statement is the one
that carries an opinion explicitly or implicitly. It is a subjective sentence. An
objective sentence contains only factual information. Subjectivity classification
deals with distinguishing between subjective sentences and objective sentences.
Sentiment classification: After applying the subjectivity classification function,
objective sentences are dropped from further analysis. Sentiment classifica-
tion focuses on subjective sentences to identify positive or negative sentiments
expressed in them. In order to identify the polarity of a sentiment, several
supervised, as well as unsupervised, learning methods can be employed. In
supervised learning based sentiment classification, Naïve Bayesian classifica-
tion algorithms, nearest neighbor algorithm, decision tree classifier, artificial
neural networks, and support vector machines are the popular methods.
For identifying sentiments, creation of a sentiment resource beforehand
is also necessary (Joshi et al. 2015). A sentiment resource is the knowledge
base that a sentiment analysis tool can learn from. It can be in the form of
a lexicon or a dataset. A lexicon is a collection of simple units such as words
or phrases annotated with labels representing different sentiments. A senti-
ment dataset is a corpus of a higher order collection of words (e.g., sentences,
documents, blogs, and Tweets) annotated with one or more sentiments. So a
sentiment resource has two components: a textual unit and labels.
Annotation is the process of associating textual units with a set of prede-
termined labels. Sentiment lexicon annotation maps textual units with labels
representing different sentiments. There are three schemes to accomplish
sentiment lexicon annotation (Joshi et al. 2015): absolute, overlapping, and
fuzzy. In absolute annotation, only one out of multiple labels is assigned to a
textual unit. The overlapping scheme is used when labels are related to emo-
tions. Multiple emotions may correspond to one positive (or negative) senti-
ment, for example, an unexpected guest or gift not only makes us happy but
surprises us as well. Emotions are more complex to represent than sentiments.
In the third scheme, a label is assigned on the basis of likelihood of a textual
unit belonging to the label. Assigning a label using a distribution of positive:
0.8, and negative: 0.2, means that the textual unit occurs more in a positive
sense but is not completely positive.
In sentiment annotated datasets, major sources are Tweets and blogs
available on social media. It includes sentence-level datasets, and document
(discourse)-level datasets. Existing labeling techniques include manual anno-
tation and distant supervision. Twitter hashtags, as provided by Twitter users,
can be used for the purpose of annotation in the distant supervision scheme.
 Data Analytics for Deriving Knowledge from User Feedback ◾ 127

An alternative to the manual approach, which is time and labor intensive, a

machine learning-based system can be trained to classify an unseen corpus
as per the observed sentiments in the training data. It predicts labels for the
words, sentences, and documents in the unseen corpus of data. The lexicon-
based methods use a predefined collection (lexicon) of words to begin with.
These words are picked while keeping in mind their ability to represent differ-
ent sentiments. Machine-learning classifiers can be trained to discover valu-
able insights in user reviews.
Mohammad (2015) refers to certain scenarios that are still challenging for
the current sentiment annotation schemes: an example is assigning a single
label to a sentence that portrays sarcasm. It depicts a positive emotional state
of mind of the speaker who actually has a negative attitude. See Mohammad
(2016) for some annotation schemes to address such issues. Furthermore,
extracting sentiments from multilingual text is another challenge at this
stage. Researchers have attempted to use the text after translating it to the
English language (Salameh et al. 2015), and the results are encouraging, pro-
vided the translation quality is good.
Sentiment composition: Subjectivity classification and sentiment classification
at the sentence level can be further aggregated to calculate document-
level classification to determine if they are opinionated or nonopinionated
(Van der Meer et al., 2011), and if they are opinionated, then to find a broad
(document-level) positive or negative sentiment. The approach is to follow
a recursive model to compose a higher order abstraction from smaller units
(Dong et al., 2014). Words when put together in an order make a sentence,
and multiple sentences constitute a document.

If a document contains a mixed response for an entity or object, then different

features of the entity or object are identified. Aspect-based techniques are used to
recognize all sentiments within a document. Positive or negative sentiments for
these features are aggregated separately. But it is not a trivial task, and demands a
computationally intensive solution.

Link Analysis
Data mining uses several techniques such as association rule mining, Bayesian clas-
sification algorithms, rule-based classifiers, and support vector machines to identify
hidden patterns in a dataset (Han, 2006). Data mining techniques, when applied
to link analysis on online social media, can provide insights into user behavior to
help policy-makers understand user feedback in a more meaningful way (Barbier
and Liu, 2011).
In social media, data mining helps to identify influential users. When a user
with a large number of followers gives negative feedback for a policy, product, or
service, the organization representatives have to be proactive to solve the problem,
128 ◾ Analytics and Knowledge Management

otherwise, it may leave a bad impression on a large user segment. The ability to
identify influential users can also help in targeting marketing efforts on people
with greater influence who are most likely to gain support for a policy, product,
or service. Therefore, it is important to understand the factors that determine the
influence of an individual in an online social network community. A simple fac-
tor can be to look at the structure of the social network that the user is a part
of. For example, a user whose Twitter account has a large number of followers,
or whose Tweets get a significant number of replies, or are retweeted widely may
indicate popularity and influence of that individual in the community. Agarwal
and Liu (2009) identify four measures to determine the influence of a blogger in a
blog community. They are recognition, activity generation, novelty, and eloquence.
Recognition is the number of inlinks to a blog post. Activity generation means
the number of comments a blog receives. Novelty follows a blog’s outlinks and the
influence value of the blog post to which the outlink points to. If the outlinked blog
post is an influential post, then the novelty of the current post is less. Lastly, the
length of a blog post determines its eloquence.
Participation of influential users in a topic increases the longevity of the topic in
the social domain. Yang and Leskovec (2011) studied temporal patterns of online
content using time series clustering. They observed that the attention that a piece
of content receives depends upon many factors including the participants who talk
about it and the topic that it relates to.
Clustering also helps to identify a community on a social network. A social
network representing a large user base can be partitioned into smaller subnetworks
on the basis of similarity of user interactions thus creating small communities of
users.
Closely related to data mining are two other multidisciplinary areas: digital
ethnography and netnography. These approaches give a scientific description of a
human society, for example, its culture or its demography. It can help to take more
informed decisions if we know the culture (integrity and honesty of the contribut-
ing users to determine authenticity of the content) of the community to which the
users belong, or their demography (young or old).

User Feedback Analysis: The Existing Work

In this section, our goal is to identify and report publications that contain a tech-
nique or approach to analyze user feedback.
Abookire et al. (2000) studied user feedback as a tool for improving the
quality of a clinic software system that was a newly implemented computer-
ized physician order entry system in 1998 at Massachusetts General Hospital,
Massachusetts. The study analyzed end-user comments regarding the software
system collected in the first year of its becoming operational. A total of 785 user
comments were classified into 18 different categories to understand the software
 Data Analytics for Deriving Knowledge from User Feedback ◾ 129

functions the users reported about. The study then analyzed the impact of this
feedback on different aspects of the software system such as efficiency, cost,
patient safety considerations, and so on. The authors opined that user feedback
can play an important role in learning and improving the system in future. In
a recent work in this domain, Vergne et al. (2013) proposed to analyze user
feedback of a software project (e.g., open source projects receive feedback from
the user community) for the purpose of understanding user requirements after
the software project is released to end users. When the feedback is analyzed and
combined with the most effective requirement suggestions for improving the
software project, it helps to identify expert users that can contribute effectively
to the requirements analysis.
Qiu et al. (2011) reported sentiment analysis dynamics of the online forum,
American Cancer Society Cancer Survivors Network (CSN), in which the users
are the cancer survivors (or their caregivers) who share their experiences of the
disease and the treatments that they took. Unlike the previous research in this
domain, this study applies computational techniques in collecting and analyzing
user responses. The dataset comprises approximately half a million posts from
the forum participants for a 10-year period of time from July, 2000 to October,
2010. The study uses machine learning classifiers to identify positive or negative
sentiments reflected in the posts. It also analyzes the sentiment dynamics and
contributing factors of this change over the period of time, that is, the change in
sentiment of a user’s posts from positive to negative or vice versa, and identifying
contributing factors, such as the number of positive and negative replies to the
post. The study reports that 75%–85% of users had a positive change in sentiment
after learning about the experiences of others with the disease and its treatment
options. However, in another study related to user feedback regarding the use of
certain types of drugs, it is found that online user reviews of the drugs provided an
exaggerated account of the impact of the drug treatment, and are much different
from the results obtained from clinical trials (Bower, 2017). This may be due to
the fact that such reviews are from people who have perhaps benefitted the most
from the drugs.
Recommendation systems also use user-provided ratings of a product to rec-
ommend a product to other similar users. Collaborative filtering is the method
that identifies the ratings that different user provide for a product. Products with
good ratings are guessed to be eligible for recommendation to other users. Another
scenario is when a user rates an item as good, other similar items (identified using
content or context filtering) are also recommended to that user. Such automatic
rating-based recommendation systems have applications in several areas including
electronic commerce and entertainment. In electronic commerce, recommending
popular items to potential buyers has been found to help in increasing sales. In the
entertainment industry, movie and music ratings on the basis of user feedback help
to recommend artists to a user on the basis of the user’s taste. Many people look for
reviews by movie critics before deciding to watch a movie in the theater. In addition
130 ◾ Analytics and Knowledge Management

to the number of reviews from movie critics for a movie, Sun (2016) takes into con-
sideration 27 other attributes such as the number of faces in the poster, and applies
data analytics to a dataset of more than 5000 movies to understand the goodness
of the movies.
It is easier to find popular items from a large inventory on the basis of user
feedback as the number of popular items is very small. However, it becomes difficult
when the inventory is too big (e.g., online catalogues) and one has to filter to get
items which obviously are not very popular (and lie in the long tail) but may be
kind of “hidden jewels” that some users may like.
As the discussion in this section shows, most of the research studies apply ana-
lytics to analyze and acquire information from data. Currently, as far as we know,
there is no link between user feedback and the knowledge management system in
an organization. A knowledge management system may support the preservation of
information acquired from user feedback as a knowledge asset, and policy-makers
in the organization can use these knowledge assets in decision-making. This paper
proposes a three-stage approach to create a link between user feedback and the
knowledge management system of an organization.

Deriving Knowledge from User Feedback

Decision-making in an organization depends upon the ability of the decision
maker to apply his or her knowledge of the past or of similar or related events to
the current situation. This ability to apply knowledge is known as wisdom, and
wisdom cannot be programmed. However, technology can be used to help the
decision maker by providing his or her knowledge regarding the past or pres-
ent events in the form of a dashboard. Knowledge is derived from information.
Information is further derived from data. Data is a collection of raw discrete facts
about certain events, which when processed (e.g., aggregated) create information.
For example, several user ideas in a user feedback system can lead to a unified idea
(see Figure 4.1).

Ideas from A unified

multiple users idea/goal

Figure 4.1 Individual ideas aggregate to a unified idea.

 Data Analytics for Deriving Knowledge from User Feedback ◾ 131

Automated data processing leads to realizing knowledge-based systems.

A knowledge-based system involves knowledge management. Nowadays, technol-
ogy is available to process large volumes of data to obtain information, to process
information to obtain knowledge, and then to put the knowledge instantly before
the decision maker for better decision-making. The previous section mentioned
automated support for user feedback analysis. However, in the context of using
user feedback to feed a knowledge-based system of an organization, there is the
need for a systematic approach to acquire user feedback and derive knowledge
from it. We propose a three-stage process to extract meaning from the mounds
of data and then preserve the knowledge for future use (see Figure 4.2). The first
step, called data management from various sources, is to acquire, store, and (clean
and) prepare the data for analysis (Labrinidis and Jagadish, 2012). Aggregating
and filtering the user data for analytics is the second step in which the data is ana-
lyzed to extract and interpret the meaning. This mountain of data contains only
1% gold; the rest of the 99% is garbage. Aggregated and filtered data here actually
refers to information.
Effective knowledge management is the third step in which we discuss which
processes should be involved so that the 1% gold is preserved and maintained for
future use. At the end, this knowledge-of-the-crowd may provide support to policy-
makers in decision-making.
We demonstrate the use of data analytics for obtaining knowledge from user
feedback with the help of a case of the Indian government in which the govern-
ment all of a sudden introduced a major change in its currency. The Indian Prime
Minister, Mr. Narendra Modi, on November 8, 2016 announced at 8:00 p.m. on
national television the immediate culling of two currency notes in the denomi-
nation of Rs. 1000 and Rs. 500. This sudden move was targeted at curtailing
money on the black market, and to putting an end to corruption in the country.
But this “demonetization” caused the general public, dealing mostly in a cash
economy, a lot of inconvenience. There was a lot of commotion on social media
regarding this. People expressed their opinions on social media platforms. In
the beginning, there were some voices in favor of the new government policy.
But with time when people had to face several hardships like long ATM lines,
and a cash crunch, they started criticizing the government. People felt that

Data
acquisition Aggregating/filtering Effective
User and data with data knowledge
feedback management analytics management
Policy
maker

Figure 4.2 Deriving knowledge from user feedback.

132 ◾ Analytics and Knowledge Management

implementing the new policy was not in the right spirit. However, for some of
them, there was no change in opinion. They appreciated the government for this
bold step.
We envisage a process model in which user feedback is first analyzed from two
perspectives—from the sentiment point of view and from the user point of view. At
a later stage, the knowledge pieces available in the information extracted from the
previous step are preserved as knowledge assets in a knowledge management system
that can further support policy-makers in decision-making. We can explain our
proposed model for extracting knowledge from user feedback using data analytics
with the help of a reference model (see Figure 4.3) in knowledge management as
suggested by Botha et al. (2008).
As Figure 4.3 shows, the user is the main focus of the model by Botha, Kourie,
and Snyman (2008). The model senses the knowledge contributed by users, organizes
it, and then disseminates the knowledge thus created using the available technology
stack. Similarly, in the model envisaged in this research, the user is the main focus.
A government organization can sense the public mood by analyzing their response
regarding a public policy on social media. Data analytics can be employed to extract
knowledge from this feedback, and then use knowledge management tools to main-
tain the knowledge assets. Lastly, the policy-makers can use and share the knowledge
repository for decision-making in the public interest.

> Human focus:

> Human focus: Organize, personalize,
Create and sense and capture

Knowledge Knowledge
creation and organizing and
sensing capture
Rich KM
solutions

Knowledge
sharing and
dissemination

> Technology focus:

Share, collaborate,
and access

Figure 4.3 The knowledge management process model. (From Botha, A. et al.,
Coping with Continuous Change in the Business Environment: Knowledge
Management and Knowledge Management Technology, Chandice Publishing,
Oxford, UK, 2008.)
 Data Analytics for Deriving Knowledge from User Feedback ◾ 133

The major challenges involved in different stages of the proposed process model
are identified as follows:

Data Management
Acquisition and recording user feedback: In this age of Web 2.0, users generate a
lot of data every second; examples are Facebook, Twitter, and RSS feeds of
news sources and blogs. Some social media sites such as Twitter and Facebook
provide Application Programming Interfaces (APIs. An example is Twitter
Auth, which helps to retrieve data from data sources. However, these sites
normally put a limit on the number of API transactions per day. For example,
for Twitter, a user cannot get history data beyond one to three weeks. And for
Facebook, the developer application can make 200 calls per hour, per user in
aggregate. This limit gives the data service companies (e.g., podargos or Gnip)
a place in the data market, and they trade data as a business entity. Given the
vast size of social media data available, an individual cannot afford to store
volumes of data even if a crawler can collect it. Moreover, a business cannot
wait for days to collect data when historical data can be purchased in minutes
or hours. Data service companies lay down the infrastructure to collect data,
and then offer data to researchers or businesses at the rate of thousands of
dollars for some months of data. Some free data sources, such as Reddit and
Kaggle, are also available.
Extracting, cleaning, and data annotation: After collecting data, some post-
processing is required to validate and clean up the data. Huge amounts
of subjective, opinionated, text documents are available in a dataset. To
improve the quality of the input, it needs to be cleaned up by removing all
the residual HTML tags; unwanted tags; stop words such as is; repeated
letters such as the “I” in hiiiii; and punctuation marks. The data is often
collected from different data sources, each of which may have the data in
a different representation and format. So, all of this data will have to be
cleaned and prepared for the process of data analysis. Annotation is the pro-
cess of associating textual units with a set of predetermined labels. Tweets
can be labeled positive, negative, or neutral on the basis of the words in the
text. There are still challenges for annotating a Tweet that portrays sarcasm.
A sample Tweet is shown below:

“Great!! When can we expect next round of #Demonetization?”

It depicts a positive emotional state of mind of a speaker who actually has a negative
attitude. Furthermore, extracting sentiments from Tweets written in multilingual
text is another challenge. Check this sample tweet:

“@narendramodi Shriman ji Jaali currency khatm hui kya... Aatankwad

khatm hua kya... Kaladhan aaya kya... #Demonetization =D>:-? O:):-/:@”
134 ◾ Analytics and Knowledge Management

A solution is to translate the text before using it. But the solution quality depends
on quality of the translator.

Data Analytics
Such large volumes of data are worthless if they are not analyzed and used. Data
analytics refers to the analysis of structured and unstructured data for the purpose
of mining meaningful information (intelligence) out of millions of entries in a
dataset. If we focus on data analytics on user feedback available on social media,
the analytics can be put into two categories: content analytics and link analytics. In
this context, content analytics corresponds to Tweet-level sentiment analysis, and
in link analysis the focus is on user-level analysis.
Various metrics have emerged to extract information from user posts.

◾ Tweet-level Analysis
– Sentiment score: Sentiment score is a numeric metric that presents the
polarity of the sentiment expressed in a Tweet. A lower score indicates
negative sentiment. The sentiment score can be further filtered on the
basis of location and gender to understand the demographic structure of
the users providing feedback.
– Understand the temporal change: The time dimension considers how fre-
quent Tweets are on a topic. Tweet frequency as a function of time may
fade with time. Content on online social media is very volatile, and popu-
larity increases in a matter of hours, and also fades away at the same
speed.
– Change in sentiment score over a period of time: User feedback can turn
from positive to negative or from negative to positive over a period of
time. Therefore, it is interesting to see the change in user response. It hap-
pened in the case of demonetization. People in favor of the move in the
beginning criticized the government when they started facing the cash
crunch, and had to stand in long ATM lines to get cash. They blamed the
government for poorly implementing the policy. Though they consider
that the move may benefit the economy in the long run, poor implemen-
tation puts them off. Sample these Tweets:

“#Demonetization All most three months passed away 90% ATM’s are
still closed as no cash, PMO must intervene and solve the issue”
“Modi is good at theory, bad in implementation. #Demonetization
effect on the poor & unorganized economy is proof ”
“How much Black Money u recovered by #demonetization? who
responsible for 200+ death, job loss, low GDP growth”
 Data Analytics for Deriving Knowledge from User Feedback ◾ 135

– Response length: User feedback in the form of longer posts may be more help-
ful. A user who spends a long time writing a response may be more involved
or have a detailed idea of the situation. But one has to look for the use of
tags, or repeated words in a post. For microblogging sites like Twitter, with
word limit of 140 characters, the response length cannot be long.
– What are the commonly used words: Use of words in a post can indicate
not only the emotions, but it can also point toward the issues affecting
the users. For example, the following Tweet highlights the status of the
cottage industry in the wake of demonetization:

“#Demonetization turned a 700-year-old cottage industry comatose.”

A word cloud can be created to indicate major issues highlighted in the

user feedback. Figure 4.4a shows a word cloud created on November 16,
2016 (available at http://rpubs.com/rajeev3176). It shows many words
such as “wedding” along with words expressing emotions such as anger,
surprise, joy, and so on. Figure 4.4b shows a word cloud created a few
months later. It shows the shift with time in the issues toward ATM and
digital or cashless transactions.
◾ User analysis: Analysis of the community structure shows links and inter-
actions in users. Link analysis can help to identify community dynamics.
The following metrics can be used to detect the community and the role of
individuals in the community:
– Popular users: This metric tells how popular the users are who are Tweeting
on the topic. Influence may depend on the number of followers of the
user. It shows the reach of the individuals in the community.
– Influential users: This metric shows the influence of the users when they
Tweet on a topic. Influence can be indicated by the number of replies and
retweets their Tweets get.

(a) (b)

Figure 4.4 (a) A word cloud on the day of demonetization; (b) a word cloud
created after a few days.
136 ◾ Analytics and Knowledge Management

– Understand the spatial change: Spatial change shows how the sentiment
was in different states in a country and in the world. People living in
small cities may have different opinions than people living in big cities.

However, the topic of the Tweet also plays a role in spreading the topic in the
community that people start talking about it. On the other hand, participa-
tion of influential users also increases longevity of a topic in the community
discussion.

Knowledge Management
The purpose of knowledge management is to ensure that the right information is deliv-
ered to the appropriate place or person for making informed decisions. User feed-
back is a valuable source of meaningful information for an organization or enterprise.
There is a need to create a knowledge management system to capture knowledge from
the information extracted from the user feedback in the previous step (explained in
the previous section) of the process. The primary purpose of knowledge management
is to support policy-makers in efficient decision-making. A knowledge base stores and
maintains knowledge assets. A knowledge management system follows standard pro-
cesses to contribute knowledge assets to the knowledge base, and to retrieve knowledge
assets from the data base for reuse in unanticipated situations.
At present, many organizations have knowledge management systems in which
employees share their experiences in a central repository. Other members of the
organization can search the knowledge base, and reuse that knowledge in other sit-
uations. For example, Infosys employees use K-shop as the online knowledge portal
available on the company intranet. Such a system is restricted to employees only.
We propose to extend it to users to incorporate their feedback in the knowledge
base. Every project should have an integrated model with a knowledge management
system at place to learn from the current implementation.
For example, in our case study in this chapter, the government can learn a lot
from the user feedback available on social media. We believe that the policy-makers
had brainstormed before introducing the demonetization move to understand the
impact of the policy on the general public. But if we look at the rule changes intro-
duced by the government after the demonetization, it seems that the policy-makers
learned a lot later when people actually started facing the problems. There were
several changes in rules regarding cash deposits, withdrawals, and exchanges until
the demonetization drive ended on March 31, 2017. Interestingly, the time period
around November 8 is supposed to be auspicious for weddings in India. Several
users mentioned the word in their Tweets on the very first day (see word cloud in
Figure 4.4). But, an exception rule was issued on November 17 (a week later) to
allow families with weddings scheduled to withdraw up to Rs 2,50,000 from their
bank accounts. It shows that policy-makers were either in a hurry or could not
foresee the hardships the genuine public was going to face.
 Data Analytics for Deriving Knowledge from User Feedback ◾ 137

This makes a case for the need of a knowledge management system as a compo-
nent of the complete process that caters to the extraction of knowledge from user
feedback, and then manages this knowledge for future use. Policy-makers can learn
from and preserve the commonly used words in the user feedback (see Figure 4.4).
In the future, when such a drive takes place again, they can consult the knowledge
base and avoid repetition of the problems in future implementations.
In addition to this, expert users can also be identified on the basis of their
involvement in topics and their understanding of the issues. When policy-makers
work on suggestions provided in user feedback for current policies or for simi-
lar policies in the future, such users can be included in the discussion group. To
select expert users as knowledge assets for a topic, there is a need to identify poten-
tial users who can contribute to that topic. For example, to discuss the impact of
demonetization on “weddings,” a set of suitable users can be identified and then
ranked on the basis of their involvement in the topic. Most suitable users can be
involved in the discussion later on.
As an example of expert users, a news editor named Narayan Ammachchi intu-
ited almost a year ago the government’s move of demonetization and Tweeted about
it. His Tweet is reproduced below (source Twitter).

“Reserve Bank of India looks to be doing everything possible to persuade

electronic payment systems such as PayTM. These are the signs that high-
denomination notes, such as Rs. 500 and Rs. 1000, are on their way out.
I have ling predicted that Modi government will use mobile payment system
to root out corruption and curb black money once and for all.”

When the demonetization actually happened, that person was recognized as an

expert to discuss the topic. Such experts can be added to the knowledge base to
invite them for suggestions on similar policies.

“@timesofindia invited our News Editor,@NAmmachchi to

@SpeakForIndia after he predicted #India’s #Demonitization a year in
advance”

Conclusions and Future Work

In this chapter we discussed a process that makes it possible to derive knowledge
from user feedback by using data analytics, and integrate that knowledge with the
decision-making process. User feedback can be a valuable resource of knowledge for
an organization. We found in the research literature that the studies in this domain
just use data analytics to extract information from user feedback, which is not
integrated with the decision-making process. Moreover, the majority of the exist-
ing work focuses on user reviews and ratings regarding electronic products, hotels,
138 ◾ Analytics and Knowledge Management

restaurants, movies, and so on. No study, to the best of our knowledge, focuses on
user feedback regarding public policies to gauge the involvement of the general pub-
lic in a government’s policy decisions.
For our analysis, we chose the Indian government’s demonetization move in
November, 2016 as a policy decision that affected almost the entire population
of the country. There was a lot of anger on the microblogging site Twitter as peo-
ple complained about the poor implementation of the policy. However, the same
government swept state elections post demonetization which indicates popularity
of the government in a country that follows a democratic system of governance.
Perhaps it points toward the digital divide in the country. Low levels of literacy
combined with a low penetration of smartphones keeps a significant portion of the
population away from the digital main stream. Furthermore, women and girls in
rural areas are deprived from using digital technology due to the dominant ideol-
ogy of patriarchy.
This study is limited to user feedback available only on the microblogging site
Twitter. However, it shows a promising direction in which work can be further
extended to integrate knowledge derived from user feedback with the decision-
making process in a public organization.

References
Abookire, S., Martin, M., Teich, J., Kuperman, G., and Bates, D. (2000). Analysis of user-
feedback as a tool for improving software quality. In Proceedings of the AMIA Symposium.
Agarwal, N., and Liu., H. (2009). Modeling and data mining in blogosphere. Synthesis
Lectures on Data Mining and Knowledge Discovery (Vol. 1), pp. 1–109. https://doi.
org/10.2200/S00213ED1V01Y200907DMK001.
Appel, O., Chiclana, F., and Carter, J. (2015). Main concepts, state of the art and
future research questions in sentiment analysis, Acta Polytechnica Hungarica,
12(3), 87–108.
Barbier, G. and Liu, H. (2011). Data mining in social media. In C. Aggarwal (Ed.), Social
Network Data Analytics, Springer, New york.
Botha, A., Kourie, D., and Snyman, R. (2008). Coping with Continuous Change in the
Business Environment: Knowledge Management and Knowledge Management
Technology, Chandice Publishing, Oxford, UK.
Bower, B. (2017). Online reviews can make over-the-counter drugs look way too effective
“Evidence-based hearsay” and personal anecdotes overshadow clinical trial results,
March 14, 2017, Science and the Public. Available at https://www.sciencenews.org/
accessed on April 14, 2017.
Dong, L., Wei, F., Zhou, M., and Xu, K. (2014). Adaptive multi-compositionality for
recursive neural models with applications to sentiment analysis. In Twenty-Eighth
AAAI Conference on Artificial Intelligence (AAAI), Québec City, Québec.
Han, J. (2006). Data Mining Concepts and Techniques. Morgan Kaufmann, San Diego, CA.
Hridoy, S., Ekram, M. et al. (2015). Localized twitter opinion mining using sentiment
analysis. Decision Analytics 2:8, 1–19.
 Data Analytics for Deriving Knowledge from User Feedback ◾ 139

Joshi, A., Ahir, S., and Bhattacharyya, P. (2015). Sentiment resources: Lexicons and
datasets. In D. Das and E. Cambria (Eds.), A Practical Guide to Sentiment Analysis,
Springer, Cham, Switzerland.
Khan, K., Baharudin, B., Khan, A., and Ullah, A. (2014). Mining opinion components
from unstructured reviews: A review. Journal of King Saud University—Computer and
Information Sciences 26, 258–275.
Labrinidis, A., and Jagadish, H. (2012). Challenges and opportunities with big data.
Proceedings of the VLDB Endowment, 5(12), 2032–2033.
Liu, B. (2011). Sentiment Analysis Tutorial, AAAI-2011, University of Illinois, Chicago, IL.
Mohammad, S. M. (2015). Challenges in sentiment analysis. In E. Cambria, D. Das,
S. Bandyopadhyay and A. Feraco (Eds.), A Practical Guide to Sentiment Analysis.,
Springer International Publishing.
Mohammad, S. M. (2016). A practical guide to sentiment annotation: Challenges and
solutions. In Proceedings of the Workshop on Computational Approaches to Subjectivity,
Sentiment and Social Media Analysis. San Diego, CA.
Pawar, A. B., Jawale, M. A., and Kyatanavar, D. N. (2016) Fundamentals of sentiment anal-
ysis: Concepts and methodology. In W. Pedrycz and S. M. Chen (Eds.), Sentiment
Analysis and Ontology Engineering: Studies in Computational Intelligence, Vol. 639.
Springer, Cham.
Qiu, B., Zhaoy, K., and Mitra, P. (2011). Get Online Support, Feel Better—Sentiment
Analysis and Dynamics in an Online Cancer Survivor Community. IEEE.
Salameh, M., Mohammad, S. M., and Kiritchenko, S. (2015). Sentiment after translation:
A case-study on arabic social media posts. In Proceedings of the North American
Chapter of Association of Computational Linguistics, Denver, CO.
Sun, C. (2016). Predict Movie Rating. Available at https://blog.nycdatascience.com/
student-works/github-profiler-tool-repository-evaluation/ retrieved on April 10, 2017.
Van der Meer, J., Boon, F., Hogenboom, F., Frasincar, F., and Kaymak, U. (2011). A frame-
work for automatic annotation of web pages using the google rich snippets vocabu-
lary. In 26th Symposium On Applied Computing (SAC 2011), Web Technologies Track,
pp. 765–772. Association for Computing Machinery.
Vergne, M., Morales-Ramirez, I., Morandini, M. et al. (2013). Analysing user feedback and
finding experts: Can goal-orientation help? In Proceedings of the 6th International i*
Workshop (iStar 2013), CEUR, Valencia, Spain, Vol. 978.
Yang, J., and Leskovec, J. (2011). Patterns of Temporal Variation in Online Media. WSDM’11,
February 9–12, 2011, Hong Kong, China.
Chapter 5

Relating Big Data and

Data Science to the
Wider Concept of
Knowledge Management
Hillary Stark and Suliman Hawamdeh

Contents
Introduction ......................................................................................................142
The Wider Concept of Knowledge Management ...............................................143
The Shift in Data Practices and Access ...............................................................144
Data Science as a New Paradigm .......................................................................146
Big Data Cost and Anticipated Value ................................................................147
Information Visualization..................................................................................148
Data Analytics Tools .......................................................................................... 152
Applications and Case Studies ........................................................................... 157
Emerging Career Opportunities ........................................................................160
Conclusion ........................................................................................................163
References .........................................................................................................163

141
142 ◾ Analytics and Knowledge Management

Introduction
The exponential growth in data and information generated on a daily basis has impacted
both personal and organizational work environments. The acquisition, analysis, and
subsequently the transformation of data into information and actionable knowledge
is dependent to a large extent on advances in technology and the development of
highly effective and efficient data-driven analytical tools. Analytical data-driven tools
have become paramount in the sorting and deciphering of large amounts of digital
information. The general consensus today is that we are living in the information age
and are overwhelmed by Big Data and the amount of information generated on a daily
basis, whether that information is at the personal or organizational level, also known
as information overload. Information overload is a reality and it has potential negative
implications both on the health of the individual and the health of the organization.
It is becoming increasingly clear that the acquisition, analysis, and transfor-
mation of Big Data into actionable knowledge will depend to a large extent on
creating cutting-edge technologies, advancing analytical and data-driven tools,
and developing advanced learning methodologies (McAfee & Brynjolfsson, 2012;
Waller & Fawcett, 2013; Chen, Chiang, & Storey, 2012). The application of these
tools will also depend on the level of investment made in training individuals as
well as organizations to manage knowledge with purpose.
Technology is the driving force behind the creation and transformation of Big
Data and data science. Today, there are many tools available for working with Big
Data and data intensive operations, but the most important and significant challenge
facing the industry is the lack of expertise and the knowledge needed for implemen-
tation. We must question, how good are these tools and who has the skillset to
operate these tools? But most importantly, who has the knowledge and expertise to
interpret the results and transform the information into actionable knowledge?
The answer to these questions depends on two types of knowledge-oriented com-
ponents that must work together in tandem. The first component is knowledge in
the form of data and information (explicit knowledge) and the second is knowledge
in the form of skills and competencies (tacit knowledge), both of which are needed
for research, development, and implementation. In the current era of Big Data and
data science-driven initiatives, many CEOs will readily attest that their competitive
advantage is the expertise of their employees, or human capital, but this in itself pres-
ents a challenge as this type of knowledge is not easily captured, stored, or transferred
(Nonaka & Takeuchi, 1995; Liebowitz, 1999; McAfee & Brynjolfsson, 2012). There
is a body of knowledge in literature concerning the relationship between explicit, or
documented knowledge, in comparison to tacit knowledge, which presents itself in
the form of skills and competencies that one would consider to be almost intuitive or
nonverbal (Hedlund, 1994; Orlikowski & Baroudi, 1991). However, simply being in
agreement that both of these components are necessary in the furthering of knowledge
will most likely not substantiate actual growth, but rather requires an active state of
ownership to transform tacit competencies to documented and shareable knowledge.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 143

The Wider Concept of Knowledge Management

Knowledge management (KM) is defined here as an interdisciplinary approach to
dealing with all aspects of knowledge processes and practices. Knowledge processes
include activities such as knowledge creation, acquisition, discovery, organization,
retention, and transfer. Knowledge practices include best practices, lessons
learned, communities of practice, learning organizations, knowledge sharing,
mentoring, shadowing, and apprenticeship. KM is concerned with the big picture
and the ability to connect the dots (Alavi & Leidner, 2001). KM introduces a
new dimension, the need for managing and transforming tacit knowledge to
documented knowledge, which can be brought about through the improvement
of communication, information and knowledge transfer, and collaboration.
One aspect of KM is creating an environment for the transfer of knowledge
to take place (Al-Hawamdeh, 2002; Hedlund, 1994; Tuomi, 1999). KM can be
viewed as the interactions of four major different components, including people,
technology, information, and processes (Figure 5.1). When each of these four
components is joined with one another, different components emerge. For exam-
ple, the by-product of the interaction of information and people is research and
intelligence activities. Information generation and gathering is a human activity
that predates civilization. When people and processes interact, it fosters infor-
mation growth in both administrative and business environments, as well as the
activities that govern many of the norms and practices within organizations. It
fosters cultures that are unique to a group of people, an organization, and society
as a whole. Combining technology-oriented processes results in the creation of

Information Analytics Technology

tools

Research and Systems and

intelligence KM workflow

Business
practices
Process
People

Figure 5.1 Knowledge management as interactions of four major components:

people, technology, information, and processes.
144 ◾ Analytics and Knowledge Management

new systems and workflows, with much of the automation coming about through
normalizing a sequence of events that have proven to be efficient and effective. The
intersection of information and technology has been the cornerstone in the way
that data and information are generated and used. It can also be viewed as the tool
needed in this case for data and test analytics. At the middle of this intersection
lie KM processes and practices needed to create the paramount transformation of
tacit knowledge to explicit knowledge. For KM to work, the various components
must be put together to enable us see the big picture and be able to connect the
dots.
Another aspect of KM is the ability to process and integrate information from
multiple sources and in varying formats, inclusive of information in such magnitude
that we now call it “Big Data,” as well as one’s reliance on data-driven analytics from
differing perspectives (Lamont, 2012). This goal is quite similar, if not synonymous,
with the idea of data science, but what is data science? Many individuals grapple
with defining data science based on its interdisciplinary nature, as it involves the
application of statistics, collection and storage of Big Data, creation of visualizations,
and so forth, but at the same time each of these items can also be housed in informa-
tion science, computer science, business, or even graphic design and art.
Data science uses automated tools and mechanics to extract and relay informa-
tion, often from very large amounts of data, to assist individuals and organizations
in analyzing their own data for better decision-making processes. These tools have
been applied to almost every realm of business and industry, as powerful insights
can be derived to then generate critical knowledge and action plans. Impacting
scientific infrastructures, workflows, processes, scholarly communication, and
certainly one’s health and well-being, the application of measurements to over-
whelming amounts of information is at the core of data science. It’s important to
keep in mind that data itself isn’t that important without a context in which to
apply it, as each organization has different needs and curiosities, meaning that a
wide variety of tools are now available to accommodate these demands for more
information.

The Shift in Data Practices and Access

It could be argued that possibly the greatest achievement in the digital revolution is
that an individual now has the ability to interact with data under their own terms
without relying on other parties to mediate access to information. Access to one’s
own analytics has given both individuals and companies the ability to use data to
measure their performance and productivity and to determine the return on invest-
ment (ROI) of various business activities. There are wide ranges of analytical tools
that have become synonymous with formal data-driven roles, but in reality, almost
all businesses and individuals rely on analytics in their daily routines in one way
or another.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 145

The stock market is a prime example of business activities that rely heavily on
analytics and analytics-driven tools. Traditionally one could not trade successfully
without relying on the knowledge and expertise of a stock broker who would have
access to information relating to trading stocks. All of the information regarding
the health of the companies that were being traded and changes within the market,
such as an initial public offering (IPO), mergers, or bankruptcy, was handled and
disseminated through a broker. With information and data about the stock market
quickly becoming available online in the late 1990s, the end user suddenly had the
ability to become an actor in what is today is termed as the knowledge economy.
The knowledge economy is a concept that revolutionizes the way we do business
and demolished the traditional geographical market boundaries. It’s difficult to
imagine a time period where we were so dependent on local markets and how
we’ve quickly become accustomed to the culture of instant access to information.
Whether it’s retrieving information via Google, the buying and selling of products
and services online, or conducting one’s banking online in real time, this depen-
dency on instant access to information is a by-product of the increased values of
knowledge, the knowledge economy, and knowledge management.
Smartphones have revolutionized the way we access information and have given
us the ability to conduct searches and perform online transactions 24 × 7, removing
the dependency on middlemen, increasing autonomy, and reducing the associated
waiting time of many transactional processes. This technology has not only reduced
one’s waiting time but has also removed many of the physical barriers that were
once in place, thereby increasing productivity. Online banking through mobile
applications is a good example, as it minimized the need to go to a physical location
to deposit checks, and provides real-time transaction history, including deposits,
withdrawals, and pending items, so that the consumer has the most accurate view
of his or her account. It not only helps the consumer feel more informed but also
assists in establishing a trusting relationship between the consumer and the bank
provider. The elements of trust and privacy were two of the main concerns to many
consumers in the past, especially in the context of online transactions, and while
this issue still remains a concern, as more and more people use online transac-
tion processes, one’s increased access to information minimizes those fears and
fosters a trusting relationship between the consumer and the brand (Rittinghouse &
Ransome, 2016; Chiregi & Navimipour, 2017).
Access to information in the medical field has been long seen as an issue of
a great concern. However, the medical field has seen tremendous change in the
last decade, with almost all hospitals and doctors’ offices transitioning from
paper charts to digital electronic records. This shift to a digital format not only
minimizes errors in record keeping, but it also allows the patient to access their
own records and take charge of their care under their own terms. Interacting
with one’s clinic as well as a pharmacy in an online environment has become
common practice. Online patient portals are becoming commonplace, espe-
cially for specialists who see the same patients on a somewhat frequent basis.
146 ◾ Analytics and Knowledge Management

This portal not only allows users to schedule appointments and find helpful
resources, but it can also work as a receptacle for storing test results, prescrip-
tion dosage amounts, and can even facilitate as a record of one’s entire medical
history. Research pertaining to patient portal use shows a significant increase in
dialog between the patient and medical staff, increased attendance in follow-up
visits, and increased recovery, with patients attributing these successes to the
feeling of being autonomous and in control of their level of care and communi-
cation (Zarcadoolas, Vaughon, Czaja, Levy, & Rockoff, 2013). Studies show that
when people feel autonomous and are made to believe that they are in control
of their choices and actions, especially when they have a medical condition that
needs their continued attention, that their adherence to a regimented protocol
improves as their access to relevant information is increased (Kipping, Stuckey,
Hernandez, Nguyen, & Riahi, 2016).
The notion of increased access to data and information is supported by the
advancement of technologies, such as the advancement of data analytics tools,
which enhances one’s knowledge and awareness. The ability to access information
at the right time within a particular situation, without the concern of one’s privacy
being violated, empowers people to make better decisions and to take control of
their lives.

Data Science as a New Paradigm

In 2009, Microsoft Corporation’s Vice President, Tony Hey, delivered one of the
first presentations of its kind, in which he referred to the current era as an age of data
science, of conducting “data-intensive scientific discovery” (Hey, Tansley, & Tolle,
2009). Hey considered data science to be the “fourth paradigm,” or a measurable
point in history in which science was taking a considerable change in direction.
He speculated that the advent of data-driven science would also usher in new theo-
ries and models, vastly impacting both the hard and social sciences, and that there
is a justified need for this area of expertise. Hey proposed three main activities
of data-intensive science to include the capture, curation, and analysis of data.
The capturing of data happens in many different ways, as one needs to already have
an understanding of the future curation needs, as this determines the collection
process (Hey et al., 2009). For example, the act of capturing data from a live experi-
ment in a lab looks very different than capturing Tweets from Twitter using an API.
The curation of data determines its longevity and future existence, and data that are
not curated is almost guaranteed to be lost. The analysis of data is a multistep ongo-
ing process, as the insights gained from the data change with each implementation.
The process is inclusive of building and utilizing databases, cleaning and modeling
the data, and data visualization.
The data-driven fourth paradigm raises questions about security, privacy, and stor-
age of data and its resulting information. People with a knowledge of cyber security,
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 147

cloud computing, and database management are highly sought after by employers, as
they want to be reassured that the business’s information will not be compromised.
Companies today value intellectual capital including the employees’ expertise, cus-
tomer profiles, and organizational memories as some of their greatest competitive
advantages (Duhigg, 2012). It’s commonplace in online sales and marketing for orga-
nizations to seek clients’ historical information, an action that is usually included in
the barely legible fine print of the contract that you agree to when making any large
purchases, such as a new car, or even signing up for a credit card (Campbell & Carlson,
2002). The compilation of this information alone isn’t of great value until one puts
forth the time-intensive effort of analyzing the data, with the purpose of making sense
and finding patterns among the records. The activity of analysis is often very time-
intensive, especially for a scientist or researcher who does not have the skillset for
the available technologies. We detail some of the currently popular tools later in this
chapter that have become mainstream in the realm of data science, particularly ones
that provide increased levels of analysis applicable to the needs of all organizations,
even those who do not see themselves in the business of data.

Big Data Cost and Anticipated Value

The terms “Big Data” and “data science” are often used interchangeably to refer to
the increased interest and anticipated value of digital information; however, Big Data
is just one component of data science. Big Data refers to structured and unstruc-
tured datasets that are large and complex (Provost & Fawcett, 2013), with one of the
main concerns of big data infrastructure being the capability to collect and analyze
large datasets beyond the traditional method of sampling and extrapolating (Walker,
2014). For data to be “big” it must meet at least four different characteristics, known
as the 4 Vs: volume, velocity, variety, and variability (Lamont, 2012; Marz & Warren,
2015; McAfee & Brynjolfsson, 2012; Russom, 2011).
Volume refers to the size of data an organization is collecting on a continuous
basis from difference sources, such as patient information, patient records, data
generated from various diagnostic machines, imaging and ultrasound devices, as
well as one’s drug history and interactions within one’s patient portal (Zarcadoolas
et al., 2013).
Velocity refers to the data streams and the speed at which data are generated,
transferred, and processed. Imagine the speed at which data are generated and
transferred using social media platforms such as Instagram and Twitter. Both of
these social media sites have the option of using a hashtag to include one’s post
within a larger but similar conversation. In the last decade, when events have taken
place that drew great attention, whether it be a presidential election or a state of
emergency, a word or phrase is turned into a hashtag, and within minutes millions
of people are connected in conversation through the use of this type of tagging. It’s
this level of speed that constitutes velocity.
148 ◾ Analytics and Knowledge Management

Variety is pretty easy to define, as it’s the inclusion of data in many different for-
mats, from traditionally structured data, such as numeric data, to qualitative data,
such as videos and audio that add complexity and robustness to one’s knowledge
of a subject.
Variability refers to the volatility in data and datasets. It is different from variety
in the sense that variability deals with the changing nature of the datasets due to
uncontrolled variables. We see changes in data over a period of time when analyz-
ing large datasets such as stock market information. The types of data collected
and used in stock market-forecasting changes daily, and is based on a wide range
of variables that include, human, social, political, and economical value. The vari-
ability of that data sometimes makes it complex and difficult to predict or forecast.
Scalability is another important area that’s necessary when dealing with Big
Data. Scalability refers to the capabilities we build to handle Big Data, from com-
puter networks, information systems, databases, analytics tools, and so on. A tool’s
scalability means that it can handle changing amounts of data depending on the
task, giving flexibility to the user to apply one technology to many situations. Some
of the popular tools for the handling of Big Data to be discussed in further sections
include Cloudera, Hadoop, and Spark.
The value of Big Data is not only in the data itself but rather in what an orga-
nization can do with it. Therefore, knowledge discovery from Big Data using
advanced data science and data analytics tools is critical to an organization’s overall
intellectual capital. But without the infrastructure and the technological capabili-
ties to make use of Big Data, questioning how to handle the data could become a
burden to an organization. Organizations will have to make significant investments
in capturing, processing, and managing large datasets, and the ROI can only be
measured by the value of the knowledge that can be harvested from that data. This
could be a challenge for an organization given the tacit nature of that knowledge.
Therefore, an organization must have an intellectual capital and KM strategy in
place that values tacit KM, and should be able to conduct knowledge audits from
time to time to justify the organization’s ROI.

Information Visualization
Data and information visualization is key to the process of making sense of big data
and data analytics. In order for the knowledge discovery from a dataset (large or
small) to be complete, there must be a definitive point of application in which the
outcomes convey meaningful action. Data itself are granular in the large scheme
of knowledge, especially if one hopes that these data will precipitate a change in
human behavior. Let’s think of this within the scope of an item that we are all
familiar with, which is the nutrition facts panel as seen on all prepackaged foods
in the supermarket. This label itself is not dramatic or enticing, probably because
we are so accustomed to these labels being present that the majority of individuals
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 149

don’t actually reference the data contained within the panel unless there is a health
concern (situation) caused by something such as diabetes or a desire to lose weight,
and so on.
The panel contains at minimum the amount of calories per recommended
serving, and the grams of carbohydrate, fat, and protein, as well as the daily
percentages needed of these macronutrients for the average healthy individual.
The visual representation of the data is determined by the frame of reference and
the understanding of the metrics used within the panel, but in order to convey
the intended message to the user, one must first have a frame of reference to
make sense of the data (Reber, 1989; Velardo, 2015). Knowledge acquisition and
utilization happen when users are able to understand information conveyed to
them in an easy manner, but very few packaged products provide visual panels to
assist users in having a frame of reference for making decisions, therefore the data
included in the panel often goes unreferenced.
While some visualization techniques are simpler to understand, such as the
nutrition facts label panel example above, not everything visual is easy to compre-
hend without a certain degree of training and expertise. Viewing medical images
without the proper training needed to contextualize the data to understand the
results can be problematic. To perform any level of analysis through visualization,
which is reported to be the most personally valued of the five senses in the Western
world (Kambaskovic & Wolfe, 2014), one must contextualize the data first and
have a degree of familiarity of the attributes of data visualization.
Data visualization is a hot topic at the moment, even commanding entire uni-
versity courses to be devoted to its discussion, as its amorphous nature includes
the joining of both quantitative and qualitative data, while dancing in the realms
of computer science, art, graphic design, and psychology (Healy & Moody, 2014;
Telea, 2014). It’s difficult to assign data visualization to one discipline as the recently
built tools rely on the finesse of computer scientists that have the skillsets to build
a program or software that renders an image based on a programmed code, such
as the outputs of Python or R, but also has a holistic presence and relies on the
expertise of artists and graphic designers. The visualization tool Tableau is espe-
cially popular in the business sector, providing aesthetically pleasing graphs and
diagrams appropriate for many different types of users, from analyzing a client’s
data to actually presenting the final analysis to the client, as its functionality and
use of color and dimension are almost limitless. The selection of colors used to relay
information shouldn’t be considered secondary to the importance of the informa-
tion, as we have been conditioned to associate colors with particular meanings or
feelings, and these associations can sometimes be more blatant than the informa-
tion itself if not appropriately assigned.
In the United States, people associate the color red with having a meaning of cen-
trality or warning, with a heat map usually assigning the deepest shade of red to the
area of most importance. When looking at a weather heat map for a storm, the loca-
tion with the most severe conditions is almost always assigned red, with the intensity
150 ◾ Analytics and Knowledge Management

fading to orange and yellow as you view the perimeter of the storm. The color red has
also been used for stoplight labeling, as a definitive warning to not proceed, and some
manufacturers have even played with using this three-color system (inclusive of yel-
low and green) to assist consumers in more easily understanding the nutritional value
of prepackaged items, such as desserts having a red label, meaning to take caution,
while milk and eggs have a green label, which gives permission to proceed in pur-
chasing and consuming. Regarding the theme of nutrition, these color-coded labels
have proven to be especially helpful for individuals in making more health-conscious
food choices, especially for those who have a limited education (including youth and
adults) and lack certain literacy skills needed to analyze the data included on the
nutrition facts label panel or list of ingredients on the back of the packaged item
(Sonnenberg et al., 2013).
Producing visualizations is especially popular as one can summarize the
importance of the information presented in a significantly faster amount of time
than it would have taken each viewer to actually comb through the data itself
in search of its meaning (Fayyad, Wierse, & Grinstein, 2002). Visualizations
by industry’s definition are simply pictorial or graphical images created for the
purpose of presenting data in a summarized format, and range anywhere from
a simple bar graph or pie chart to a 3D color-saturated heat map with mov-
ing elements (https://www.sas.com/en_us/insights/big-data/data-visualization.
html). With this definition in mind, we challenge the reader to consider all of
the visualizations that one encounters on a daily basis, as we take many items for
granted in our quick analyses that represent quite complex amounts of informa-
tion. One of the most popular visualizations that summarizes a large amount of
data into one single image, is Napoleon’s successive losses as his French troops
crossed Russia from 1812–1813, as drawn by Charles Minard in 1869 (Tufte,
2001; Friendly, 2002). Minard is considered the first to use graphics in combina-
tion with engineering and statistics, specifically numeric data as represented on
graphical maps.
The map below combines six different types of data: the number of soldiers
within the troops; the direction of the travel (it changes as they advance and
then retreat); the distance traveled across Russia; the temperatures encountered;
latitude and longitude; and the location of each report. The thick beige line
represents the amount of soldiers within the troop at the start of their journey,
which is seen to diminish as the line extends from left to right. The black line
then represents the amount of soldiers that were still alive as they then began
the second half of their journey, which was to return to the western-most point
of the map, as seen when the black line diminishes from right to left. It’s evident
that based on the scale of these two lines, the majority of the troops did not
complete the entire march across Russia and back again. This image is often
included within the coursework of visualization classes, as it compels the viewer
to consider all of the different aspects and the amount of information included
in such a simple graphic.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 151

Minard’s carte figurative of Napoleon’s 1812 campaign. (From Tufte, E., The
Visual Display of Quantitative Information, 2nd ed., Graphics Press, Cheshire,
CT, 2001.)

Infographics, or a graphic made with the goal of relaying a large amount of

information, usually presented in an aesthetically pleasing format, have become
increasingly popular over the last few years. Many businesses have jumped on
board in utilizing this modern format of relaying large amounts of information to
their consumers, that most likely took the team of data collectors a significantly
longer amount of time to dig up to create the graphic, than the actual production
of the graphic itself. This type of content has proven to be especially useful when
it comes to search engine optimization (SEO), as users process visual information
much more quickly than text and are infinitely more likely to share visual content
across their social media platforms (Stark, Habib, & al Smadi, 2016). However,
consumers should be cautious of the quality and accuracy of the information
provided.
Tufte (2001) warns of the “lie factor” often included in graphical imagery,
which is the value used to describe the relationship between the size of effect
shown in a graphic and the size of effect shown in the data, as this is often
misinterpreted when purely shown in graphical form. Research pertaining to
one’s perception of the scientific validity of health claims made on medicine
packaging shows significance in the pairing of the textual claim with a data-
driven graph or image that supports the theme of the claim (Tal & Wansink,
2016). We do need to be wary of these visual techniques though as they are not
always founded on actual data or statistics, but instead lend a scientific tone to
the manufacturer’s claim, such as including an image of a chemical formula,
even if this is not relevant to one’s information needs, such as when shopping for
Tylenol or aspirin.
152 ◾ Analytics and Knowledge Management

Data Analytics Tools

Considering the magnitude of data currently collected and made available, it’s prac-
tically impossible to manually process that much data in the hope of finding trends
or certain patterns that might guide the decision-making process. It is much easier to
read every answer collected by a 20-question survey with 100 respondents than it is
to comprehend the same survey with 10,000 or 1 million respondents. To make the
point here, as the amount of data increases there is a need for an automated means
of analyzing data. Proprietary hardware and software provided by some of the big
players in the data-wrangling arena include but are certainly not limited to IBM’s
Watson Analytics, SAS Enterprise Miner (SAS EM), and Tableau (Lamont, 2012).
We discuss these three tools as they are marketed towards both business and
academic sectors, as well as introduce popular open-source coding languages that
can be used as alternatives. An entire book could be devoted to listing all of the
technologies and tools that have been created to assist in the mining and analysis of
data. That being said, we believe that anyone with an appreciation for data analytics
or a desire to improve upon their future knowledge acquisition, should be familiar
with the handful of currently popular tools that we detail in this section.
Watson Analytics, not to be confused with the supercomputer that became
the ultimate winner of the popular TV show Jeopardy! in 2011, is a suite of
solutions offered to businesses of all sizes by the powerhouse IBM. Trained to
use artificial intelligence (AI), Watson Analytics relies on a cognitive computing
system and is adept enough to sense, predict, and even infer solutions based on
the imported data and given conditions (Guidi, Miniati, Mazzola, & Iadanza,
2016). What differentiates this product from its competition is its functionality
to automatically analyze datasets, measure and examine the quality of the data,
and recommend the optimal statistical solution to the query. Built using natural
language processing (NLP), users can request and receive answers in almost real
time in the format of exploratory, predictive, and visual analytics (O’Leary, 2017;
Hoyt, Snider, Thompson, & Mantravadi, 2016). The amateur user shouldn’t be
intimidated by the idea of learning to use a tool such as the one described above.
The advent of digital analytics-driven technology fosters the mindset that prod-
ucts created must be flexible and approachable, growing with the user, and not
outpacing or lagging behind (Raskin, 2000).
The tool is available through partnership with IBM free of charge to academic
institutions who do not seek to commercialize their findings, including both fac-
ulty and students. Watson Analytics can be used as a resource in the classroom in
which students can try to familiarize themselves with its functionality. As Watson
Analytics only came to the market in 2015, we are now beginning to see academic
case studies being published pertaining to the real-life application of this tool
(Guidi et al., 2016). For example, the exploratory and visual analysis functions
were recently harnessed in the medical field to assist in the identification of heart
failures, through the measurement and analysis of electrocardiograph signals.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 153

SAS EM is powerful data mining software, and the makers claim that it offers
more predictive modeling techniques than any other currently available data min-
ing or data analytics tool in the market. Within the interface, models are built with
a visually pleasing process flow diagram environment, giving users the ability to test
the accuracy of their predictions using graph-oriented displays. The appeal to users
in both business and academic sectors is that the software is scalable, meaning that
it can be utilized by an individual as well as an entire team of people, can operate
on datasets of varying sizes, and that it is available in a cloud-based environment.
Built on the SEMMA methodology, which stands for sampling, exploring, modi-
fying, modeling, and assessing, SAS EM provides users the ability to find patterns
and relationships in Big Data. With user-friendliness always being a concern, SAS
EM offers a clean interface simple enough to not overwhelm an individual with
minimal statistical knowledge, but complex enough to still be of use to someone
well-versed in modeling, designing decision trees, neural networks, and performing
either linear or logistic regression (Matignon, 2007).
In 2008, researchers used the neural network method of SAS EM to diagnose
heart disease, based on the patient profiles of individuals with and without heart
disease, and produced an 89% accuracy in their predictions (Das, Turkoglu, &
Sengur, 2009). This ensemble-based methodology was made possible by splitting
the dataset into two parts, one containing 70% of the data for testing, and the
remaining 30% for validation of the proposed system, which is an integral part of
neural network modeling, as it’s impossible to confirm the accuracy of a model if
it’s not applied to data similar to what was used in its creation.
Visualizations have come a long way in a very short amount of time, and com-
bine the resources of art and graphic design with analytics, to create concise images
that represent much more complex amounts of information (Fox & Hendler, 2011).
For example, almost all of the Fortune 500 companies utilize infographics, which
are visual images that typically include text, charts, and diagrams. Tableau uses
drag and drop dimensions and offers users simple data mining capabilities that have
historically only been available to those with a complex coding skillset (Tableau.
com, 2017). Tableau gives users the ability to create time series-driven data, such
as a moving depiction of the path of Hurricane Katrina across the Gulf of Mexico
in 2005. Based on the data provided, the system was not showing only the move-
ment but also the change in intensity of the storm through the use of color as well
as varying degrees of opacity. Users can easily use the software interface to create
visualizations and save and share the results in a sharable format, such as a PDF,
JPG, or PNG.
To spark interest and plant a seed of inspiration, Tableau’s website offers thou-
sands of examples of visualizations that have been built using real-world data
examples. To incentivize educators to use Tableau within their classrooms, instruc-
tors as well as students are given a free annual license, and instructors have the
added benefit of access to course materials created by Tableau that can be integrated
within their own course research objectives. Librarians at Ohio State University
154 ◾ Analytics and Knowledge Management

(OSU) utilized Tableau to visualize library data on three very different datasets,
all of which produced results that better enabled them to meet the needs of their
university patrons. The datasets illustrated the library’s potential to promote library
collections in tandem with marketing the library’s programming, how combining
data from multiple sources could provide support for the prioritization of digitizing
special collections, and lastly the ability to better understand user satisfaction from
survey data through the Tableau’s filtering functions (Murphy, 2015).
While these tools are considered to be cutting edge, their accessibility comes at
a cost, most commonly seen as an annual licensing fee either by person or institu-
tion. For parties who desire the same results, such as text analysis and data mining,
but don’t have the resources to purchase the aforementioned software, there are a
number of free alternatives. Users should be reminded however that anything adver-
tised as being “free” usually has hidden costs, an idea popularized with the acronym
TANSTAAFL, meaning “There ain’t no such thing as a free lunch” (Grossmann,
Steger, & Trimborn, 2013). Within the context of tools and technologies, this
cost is invariably the user’s time and attention required in the beginning, to learn
how to navigate the tool and generate the desired results. A prime example of this
would be the open source and very popular coding languages of R and Python.
Python is a multifaceted language and offers everything from data cleaning to graph-
based visualizations, with over 40% of data scientists reporting it as their “go-to” for
problem solving, per the 2013 survey by O’Reilly (http://www.oreilly.com/data/free/
stratasurvey.csp). It’s really no wonder why this language has been so widely accepted,
as it supports all kinds of data formats, from CSV to SQL tables, and the online
community is robust in offering examples and inquiry support (Perkel, 2015). R on
the other hand, while also open source, is considered by many users to have a steeper
learning curve than Python, due to its complicated syntax, but it is built with the data
scientist in mind, as its language is focused on solving statistical problems and generat-
ing scientific visualizations (Hardin et al., 2015; Edwards, Tilden, & Allevato, 2014).
The results provided by writing code from either of these languages can assist the
user and spark hidden discovery, but to arrive at this point, the user will first need to
spend a great deal of time acquainting themselves with the syntax for both of these
languages. Almost all undergraduate computer science programs offer courses in pro-
gramming languages, with Python currently being the most popular introductory
language offered at colleges within the United States, compared to business schools,
which advertise their access to the proprietary tools mentioned prior (Brunner &
Kim, 2016). As discussed more deeply in other sections of this chapter, there is an
apparent disconnect in the data analytics-driven tools introduced to students within
the collegiate environment depending on their registered course of study, which lends
these areas of expertise to being siloes rather than shared for the mutual benefits of all
parties who have the same interests (Thompson, 2017).
Additional platforms and tools for the mining of large datasets and creating
models include Cloudera, and Apache’s Hadoop and Spark. Hadoop has made
its name in the software arena in two ways, through its unconventional storage
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 155

capabilities (splitting files into large blocks that are then distributed across multiple
storage systems, with the data being processed in parallel across multiple nodes), as
well as its lightning-fast processing procedures that are built upon the MapReduce
framework (Padhy, 2013). Similar to Hadoop, Spark can handle large datasets and
its speed is marketed as being up to 10 times faster than that of Hadoop. Both of
these tools have a learning curve, but are considered to be some of the more user-
friendly options for tackling gigantic datasets using open-source packages. These
tools are useful to companies that have large and complex datasets, and can also
be used in tandem, as they are not mutually exclusive. It depends on the size of the
dataset and the needs of the end users as to which option is more appropriate at that
time (Lin, Wang, & Wu, 2013).
A precursor to the age of Big Data, was the advent of e-commerce, or the selling
of products and services online, which we can see as early as the late 1990s (Kohavi,
2001), and it was in this online environment that the transactional process was
made completely trackable. Google has also made this tracking process too easy to
not take advantage of, especially for small businesses with limited budgets, through
its free Google Analytics dashboard (Clifton, 2012). To benefit from the bevy of
data available for collection through Google Analytics, snippets of code are placed
on each page of the specified website, and within seconds, the user can monitor
the engagement, activity, and trends forming based on visitors to the site (Fang,
2007). These analytics are available in real time in an interactive dashboard that is
extremely user-friendly, and can even be integrated through APIs to populate other
third party dashboards. While there are many other offerings in the Google suite
of services, we focused on this one item due to its market share, which is estimated
to having been used in more than 90% of the websites available today.
In 2008, Google launched an effort to help predict trends in illness, specifi-
cally the flu, based on users’ search queries. The Center for Disease Control (CDC)
has a lag time of up to three weeks in collecting data and then making the formal
announcement of an epidemic, but Google was able to predict trends in almost
real time through their data collection mechanisms (Lazer, Kennedy, King, &
Vespignani, 2014). When initially introduced, the idea that illness and disease had
the potential to be minimized through pinpointing the volume and location of
searches based on similar symptoms, the announcement was met with great fanfare
and excitement. This venture gave hope to not only the medical communities in the
United States, but to much poorer countries as well, where medical resources are
often limited and time is of the essence.
However, in August 2015, Google quietly terminated this initiative, due to an
increasing amount of inaccurate predictions, as Google’s model failed to adequately
account for changes in user search behavior (Pollett et al., 2016). As users, we’ve
become much more adept and accurate in our search queries, as the popular search
engines’ algorithms now all rely on aspects of semantic analysis, or the meaning
and context behind the entire search phrase, rather than just the meaning of each
individual word used within a phrase. This improvement in user search behavior,
156 ◾ Analytics and Knowledge Management

coupled with the fact that not everyone who was searching using terms that are con-
nected to having the flu, actually had the flu, produced inaccurate results based on
the model built. Someone may have searched based on having a simple headache,
while another user may have entered a query to learn more about flu season itself,
or where to get a flu shot in order to be preventative. Researchers are hopeful that in
time the model first introduced will be finetuned and be able to deliver an accurate
analysis and prediction based on real-time search data as originally promised.
While not normally called “data analysts,” those who monitor any kind of tech-
nology that records data to be retrieved and used at a later date, certainly utilize
a similar skillset. The world of medicine and medical devices has certainly been
positively impacted by the integration of analytics, namely digital analytics, as
healthcare professionals now have greater precision and clarity to provide the safest
and most efficient care possible (Alyass, Turcotte, & Meyre, 2015). There is a long
list of technologies that have transformed the way we conduct business, but the
foundational element to all of these tools is that the user is now in control of being
able to generate new information that will in turn become knowledge, and through
application, the promise of progressive change is possible (Howe et al., 2008).
Every aspect of business has been impacted by increased access to data analyt-
ics, even including the wine industry, which has traditionally been perceived as a
more manual field, steeped in the romanticized idea of picking and sorting grapes
by hand to preserve a certain level of product quality (Dokoozlian, 2016). Many
wineries are starting to integrate optical sorting machines in their facilities, the
leading branding currently being WECO (http://www.weco.com), which operates
based on a camera with LED lights that illuminates each individual grape during
the sorting process, measuring the fruit’s surface for defects as well as sorting based
on the desired size of the grape, and discarding all of the items that don’t meet the
set standards for optimal wine production, including stray twigs, leafs, and debris.
The system can even be calibrated for color and density, selecting grapes that
have particular hues that represent the ripeness or saturation of sugars within the
grape, as each wine maker desires different levels of these components. While one
may argue that this is a stretch for an example of data analytics, analytics themselves
don’t have to be a physical printable image or graph, but rather information that is
derived from the application of measurements or statistics. These same machines
are also used within the farming and food manufacturing channels, revolution-
izing entire industries through minimizing manual labor and risk while increasing
accuracy for a wide gamut of items that we consume on a daily basis, including
but not limited to nuts, olives, berries, and grains. This type of optical sorting has
been used for some time now, specifically within the grain industry, as a method of
removing grains that are infected or contaminated, minimizing the risk of illness
in both humans as well as livestock (Dowell, Boratynski, Ykema, Dowdy, & Staten,
2002). While a data-driven tool may have been originally created to facilitate the
needs of a more traditional or scientific disciplines, with time, most industries have
also started to integrate and adapt these practices.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 157

Applications and Case Studies

The increased efficiency, transparency, and rapid growth of information can be attrib-
uted to advances in science and technology including analytics and KM. For example,
it’s safe to assume that the majority of World Wide Web (WWW) users have used at
least one of Google’s products, or are at least familiar with the purpose of their offer-
ings, such as the Google search engine, Google Earth, or Google Maps (Jarvis, 2011).
Available for downloading on a smartphone, the majority of the users of this applica-
tion rely on its precision and accuracy for producing a visualization of a map of the
most efficient route to take at that time, inclusive of variables such as traffic or road
construction. We take for granted the integration and magnitude of data necessary to
facilitate this application running efficiently, which is inclusive of thousands of data
points all being evaluated and analyzed in tandem, to then produce multiple options
for our desired route—such as tolls or no tolls, or if there is a coffee shop on the drive.
The data collection points utilized by Google are more recently also being used by
local governments, specifically in the endeavors of smart cities (Vilajosana et al., 2013).
“Smart cities” is a relatively new and still unfamiliar term to many individuals out-
side of the realms of technology and city governance, but one of the primary con-
cepts is that data is constantly being collected that could be of greater value if made
easily available to both city authorities as well as the citizens of each municipality.
Communities of varying sizes are advocating for the collection and transparency of
this data to be made available, as well as increased analytics generated to provide safer,
healthier, and happier environments for their citizens. An example of a large metro-
politan that is considered a smart city and provides resources built of off analytics-
driven tools to its citizens, is the city of Chicago. Connecting over 250,000 street light
fixtures, Chicago’s smart light project is believed to be the largest city-wide wireless
smart light program in the United States, and is built using LED lights that can be
remotely dimmed or brightened as needed.
The network can also remotely monitor lights in need of repair, sending an
alert instantaneously after a failure has occurred, and is integrated into Chicago’s
non-emergency 311 system (http://www.chicagoinfrastructure.org/initiatives/
smartlighting). At a smaller scale, the City of Denton, Texas, has revamped their
waste-collection services through increased access to municipal data. By adding sen-
sors that monitor the fill-level of each large dumpster located in the downtown area,
the city can track in real time the need for increased collection and disposal, and
design their workflow routes and employee schedules to meet the needs of the city.
This increased access to data provides not only information that can make the
city more efficient in their operations, but also provides a cleaner more sanitary
environment for its citizens who frequent the area. Their future “smart city” projects
include citizen access to utility consumption across the city, including water, gas,
and electricity, to have an increased awareness of the resources used as well as to
be able to be more efficient in their consumption, such as not overwatering public
green spaces in the summer or reducing the electricity used in unoccupied buildings.
158 ◾ Analytics and Knowledge Management

Even if you don’t live in a smart city (yet), you probably go to the grocery store,
or have a membership to one of the bulk-item retailers, such as Costco or Sam’s
Club, and the chances are pretty good that you have a loyalty card tied to one of
these merchants, either physically presented or keyed-in using your name or phone
number at the point of transaction. Users are generally incentivized to use these
cards to receive discounts, and at the same time the retailer collects data on which
items were purchased together and at what frequency, and so on. This process has
become the norm and most of us no longer give the sharing of this information
a second thought, but what if the data contained within these historical records
could actually be of a much more beneficial use than originally estimated?
A prime example of harnessing this innocent, some would even say unneces-
sary, data for knowledge utilization and the betterment of humanity, is the research
conducted at the University of Oxford, with a keen interest in both Big Data and
dementia (Deetjen, Meyer, & Schroeder, 2015). Dementia is a good example of a
chronic disease that often causes an individual to exhibit particular behaviors for
quite a long period of time before diagnosis, and as many chronic diseases, it is
speculated to be heavily influenced by diet.
The researchers posit that medical professionals should utilize grocery store
loyalty card data to have more information regarding the prior dietary choices of
individuals who have a current diagnosis of dementia, as well as to be proactive in
spotting the early signs of this disease in its first stages. One of the behaviors exhib-
ited in the early stages of dementia is that of forgetfulness or decreased memory,
and this is especially prevalent when grocery shopping, with some people buying
the exact same item over and over again with increased frequency. Multiple grocery
store brands have already volunteered to share this data with the medical profes-
sion, and if approved by the legislation, the release of this longitudinal data for
analysis could prove to be quite life-altering.
Another topic of recent debate is the 2016 United States presidential election.
The outcome of the election surprised both analysts and statisticians who analyzed
and presented data, which led the majority to believe that the Democratic candidate
would win by a landslide. A popular website predicting the outcome of the election
was http://www.FiveThirtyEight.com, hosted by Nate Silver, who used vast amounts
of data collected, paired with rigorous methodology, to provide a quantitative analy-
sis of the discussions and opinions heard across a broad audience. They correctly
predicted the 2008 election won by Barack Obama, with his final presidential elec-
tion forecast accurately predicting the winner of 49 of the 50 states, missing only
Indiana, as well as the District of Columbia. He again shocked the nation by accu-
rately predicting the outcomes of all 50 states in 2012, including the 9 swing states,
but then in 2016 the model failed. They predicted that Hillary Clinton would win
by a 71% majority vote, giving Donald Trump only 29% of the vote.
The data analyzed included a wide range of polling methods, a similar approach
to the data collected for the two previous elections. A reason given for this error
in accuracy was not the method used, but rather in the reliance on the final data
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 159

collected, an example that the variability of Big Data could change the predictive
outcome. Social media platforms played a large role in this false feeling of security,
as users’ posts so often lack independence in this age of digital transparency, and
while a single Tweet can be viewed hundreds of thousands of times, it really only
represents the sentiment and opinion of a single person (Rossini et al., 2017).
The election has brought increased attention to the information gathered, and
subsequent knowledge formed, by users who rely solely on data presented online,
specifically from social media sites such as Facebook and Twitter.
Social media as a channel for communication has reshaped the way that we
form our social circles, with conversations never actually dying as online banter
and behavioral data are constantly tracked and stored. Businesses are encouraged to
take advantage of the immense amount of data that we have chosen to share about
ourselves through our social media profiles to improve their marketing strategies,
with the most popular and robust platform currently being Facebook. The behav-
ioral tracking and data collection efforts of Facebook include behaviors online and
offline, making the connection between user profiles by using credit card informa-
tion as well as email addresses and phone numbers (Markovikj, Gievska, Kosinski, &
Stillwell, 2013). Advertisements can then be tailored and shown just to the intended
online audience, saving the business both time and financial resources, while maxi-
mizing their ROI.
Researchers and academics are also benefiting from stored information as it pro-
vides a more concrete method of discovering communicative patterns exhibited across
networks, including informal peer-to-peer interactions as well as business to business
(B2B) or business to consumer (B2C) data. Twitter data has been especially useful for
having a deeper understanding of users’ thoughts and opinions, through the inclu-
sion of hashtags, which note the meaning of the conversation and connect users of the
same verbiage to one another. For example, if I tweeted about the 2016 presidential
election and used one of the president campaign hashtags, such as #MAGA (Make
America Great Again), my Tweet would then be added to the list of all other Tweets
that used the same hashtag, essentially creating a conversation around this topic. It
is also possible to download all of the Tweets using this hashtag and perform a senti-
ment analysis to determine the positive or negative feelings associated with this topic.
Twitter’s API allows for batch downloads of Tweets, currently up to 10,000 or
more Tweets at once, and includes the user’s location as well as the dissemination
and breadth of share across the Twitter network. Technologies are abundant in
fulfilling the need to visualize this type of data. Within the context of the afore-
mentioned example of downloading Twitter data, NodeXL could be an appropriate
tool for visualization as it provides a visual representation of the network breadth
and sharing of a Tweet or hashtag topic, and is user-friendly as it’s an Excel plugin,
found at http://www. NodeXL.com. This plugin also provides a list of the top 10
influencers within these conversations based on their centrality, including likes and
retweets, as well as the top URLs included in the Tweets, making it much more
evident as to who is leading and participating in certain conversations.
160 ◾ Analytics and Knowledge Management

Emerging Career Opportunities

One of the notable effects of the data science and data analytics revolution is that
new career opportunities are emerging. The job of “data scientist” was termed the
sexiest job of the twenty-first century, although it is certainly not the easiest, as
employers want their data analytics-focused applicants to wield a set of skills much
akin to that of a Swiss army knife (Davenport & Patil, 2012; De Mauro, Greco,
Grimaldi, & Nobili, 2016, 2017; Gardiner, Aasheim, Rutner, & Williams, 2017).
Not only must a prospective applicant first be able to wrangle large amounts of
data, often in multiple formats that don’t integrate with one another easily, but
after the task of cleaning, mining, and modeling the data, they must then be able
to glean the important pieces of information hidden within the data and infer its
application including meaningful presentation of the results.
Since this is a leading area of job growth, universities are racing to develop
coursework and programs to attract and educate potential future data scientists.
Given that data science and data analytics bridge many academic fields, such as
computer science, engineering, information science, business, and even art and
design, most of the programs offer an adopted interdisciplinary approach. Some
of the universities created interdisciplinary programs where students are taking
coursework across a multitude of departments while others house the entire pro-
gram within just one department. Given that this area is still considered to be new
in the grand scheme of academia, there is not yet a benchmark of classes or clear
curriculum (Tang & Sae-Lim, 2016; Lyon & Mattern, 2017; Tenopir et al., 2016),
and as a result, graduates from these programs will have a varying degree of compe-
tence in many areas often associated with the title of “data scientist.”
Some of the academic institutions that offer programs in data science and
data analytics include Massachusetts Institute of Technology (Master of Business
Analytics), Carnegie Mellon (Master of Information Systems Management: Business
Intelligence & Data Analytics), University of Texas at Austin (Master of Science in
Business Analytics, Master of Science in Computer Science), University of Chicago
(Master of Science in Analytics), Rutgers University (Master of Science in Data
Analytics), Southern Methodist University (Master of Science in Business Analytics),
and University of North Texas (Master of Science in Data Science; Master of Science
in Advanced Analytics). We compare the programs offered at these institutions to
find commonalities as well as what differentiates, depending on if a student enrolls
in a more business versus computer science-oriented program of study.
The similar concepts taught across the programs include students gaining a
working knowledge of database management, data mining, data visualization,
the application of tools used to collect, store, transform, model, and analyze large
amounts of data (tools specifically mentioned within their curriculum descriptions
include but are not limited to SAS EM, SPSS, Hadoop, and Tableau), marketing
analytics, and the ability to apply a wide range of statistics to perform scientific
analysis. Most of the programs offer both full-time and part-time options. They
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 161

acknowledge that many of their students will be at least partially employed dur-
ing the time that they complete their master’s degree, and need the flexibility in
the amount of coursework taken each semester. Many of the degrees also include
completing an internship or team-based project that is then presented to a panel
of executives, greatly emphasizing the importance of communication and team-
oriented skills within the workplace.
Positions that one can apply for upon earning their Master of Business or
Science degree with an emphasis on data analytics, are on the rise as more and more
companies are creating the roles of data scientist, manager, analyst, consultant, or
software or project engineer, in order to stay competitive (De Mauro et al., 2016,
2017; Gardiner et al., 2017). Upon review of some of the job postings that included
the keywords “data analytics,” “data analyst,” and “data scientist” on three popular
recruitment sites Glassdoor.com, LinkedIn.com, and Indeed.com, we identified a list
of the common minimal qualifications desired by employers for their future hires:

◾ Bachelor’s degree in mathematics, engineering, computer science, or statis-

tics, with 3 or more years of work experience or academic research.
◾ Proficient in more than one statistically driven coding language, specifically
R, Python, and SQL.
◾ Experience with modeling and visualization software, specifically SPSS, SAS,
and Tableau.
◾ Ability to work individually as well as be a team player.
◾ Strong written and verbal communication skills, comfortable giving techni-
cal presentations.
◾ Strong attention to detail, ability to solve complex problems, has a strong
sense of curiosity.

Optional but preferred skills that help candidates stand out from the crowd:

◾ Master’s or PhD in the aforementioned disciplines.

◾ Proficient in core programming languages, including C/C++, Java, Scala,
and Ruby.
◾ Experience using Apache’s Hadoop and or Spark to handle and process very
large datasets.
◾ Machine learning techniques, such as logistic regression, decision trees, ran-
dom forests, neural networks, and naïve Bayes.

Access to particular tools learned in academia differ greatly by employer and it’s impor-
tant that students realize that although they may be proficient in one software, that
the possibility is almost guaranteed that they will need to learn new tools and new
systems upon employment. For example, it’s rare that a small employer who is in the
beginning stages of handling large datasets would have the budget for a yearly Tableau
or SAS license, and instead prefer relying on free or budget-friendly resources. In this
162 ◾ Analytics and Knowledge Management

case, the open-source items discussed prior would most likely be the more attractive
option; therefore, a data analyst must be agile and flexible in learning new tools that
are similar to ones having been taught within the classroom. As is with any technology,
the industry that makes these data-driven tools is quickly advancing and products are
constantly being upgraded. This means that a data analyst must be open to learning and
keeping up with the new market trends (McAfee & Brynjolfsson, 2012).
Regarding the desired experience, students should utilize internships and become
involved in research projects that utilize real datasets that would be of interest to future
employers, to establish a portfolio of work experience (http://data.gov). Certifications
can also add credibility and enhance one’s resume, especially for programs or software
that have multiple levels of proficiency. For example, a leading online instructor, Lynda.
com, offers courses and tutorials in learning Python, while Tableau holds a tighter rein
and true to its proprietary nature, is the only one to have ownership to offer the two-
level certification for its software. Both of Google’s Analytics and AdWords platforms,
which measure one’s proficiency in analyzing a website’s online presence, health, and
the process of running an online advertisement campaign, have multiple proficiency
exams, but to stay active and up-to-date, these must be retaken every 18–24 months.
And while no one really enjoys having to be recertified on a continual basis, Google
has incentivized users to stay current by eliminating the certification fee, making all of
the exams now free. Coursera.com offers short online classes that cover many of the
concepts considered to be part of data analytics, with the lessons often being offered
free of charge, giving the users the option to pay for the course if they want to take the
exams and earn a certificate of completion.
Data-related internships, boot camps, hackathons, and incubators are other
options to gain relevant real-world experience outside of the classroom (Anslow,
Brosz, Maurer, & Boyes, 2016). For example, the Data Incubator program is
funded by Cornell’s data science training organization (http://thedataincubator.
com), and offers a free advanced 8-week fellowship for PhD students and grad-
uates who are about to start searching for industry positions. Companies that
have partnered with the Data Incubator for their hiring needs include LinkedIn,
Genentech, Capital One, and Pfizer, with alumni now employed by companies
such as Verizon, Facebook, KPMG, Cloudera, and Uptake. This fellowship is
typically 8 weeks in length, with time requirements similar to that of a Monday
to Friday, 8 a.m. to 5 p.m. job, and are physically held in the New York City,
San Francisco, Seattle, Washington DC, and Boston metros. There is also an
online option that typically requires participants to dedicate 2–3 months of time,
but at a part-time commitment. Technical training curriculum in the fellowship
includes software engineering and numerical computations, NLP, statistics, data
visualization, database management, and parallelization. The soft skills gained
from the training include ramping up one’s communication skills, as academics
and those in industry communicate in very different ways, along with face-to-
face networking, and practice interviews to assist fellows in being the most pre-
pared when applying for jobs upon completion of the program.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 163

Conclusion
This chapter served as a review of some of the pertinent areas of analytics (data science,
data analytics, and Big Data) and related them to the wider concept of KM. The value
and importance of these concepts is linked directly to the increased shift in the econ-
omy and the increased emphasis on knowledge and intellectual capital as a key driver of
the knowledge-based economy. The currently available tools and the applications that
we’ve outlined will soon be considered ancient history, but the fact remains that the
acquisition of knowledge and its systemic management will continue to challenge both
scientists and practitioners. We would be remiss to not stress the increasing importance
of research and academia and their role in creating more consistent theories, models,
and frameworks to guide the development and implementation of data science and data
analytics. It is also important to create a curriculum to support knowledge for students
who desire to be part of the ever-present data and digital revolution. The task of manag-
ing knowledge effectively is built on the collective wisdom and knowledge of each of the
contributing people within the organization.

References
Alavi, M., & Leidner, D. E. (2001). Knowledge management and knowledge manage-
ment systems: Conceptual foundations and research issues. MIS Quarterly, 25,
107–136.
Al-Hawamdeh, S. (2002). Knowledge management: Re-thinking information manage-
ment and facing the challenge of managing tacit knowledge. Information Research,
8(1), 143.
Alyass, A., Turcotte, M., & Meyre, D. (2015). From big data analysis to personalized medi-
cine for all: Challenges and opportunities. BMC Medical Genomics, 8(1), 33.
Anslow, C., Brosz, J., Maurer, F., & Boyes, M. (2016, February). Datathons: An experience
report of data hackathons for data science education. In Proceedings of the 47th ACM
Technical Symposium on Computing Science Education (pp. 615–620). ACM.
Brunner, R. J., & Kim, E. J. (2016). Teaching Data Science. Procedia Computer Science,
80, 1947–1956.
Campbell, J. E., & Carlson, M. (2002). Panopticon. com: Online surveillance and the com-
modification of privacy. Journal of Broadcasting & Electronic Media, 46(4), 586–606.
Chen, H., Chiang, R. H., & Storey, V. C. (2012). Business intelligence and analytics: From
big data to big impact. MIS Quarterly, 36(4), 1165–1188.
Chiregi, M., & Navimipour, N. J. (2017). A comprehensive study of the trust evaluation
mechanisms in the cloud computing. Journal of Service Science Research, 9(1), 1–30.
Clifton, B. (2012). Advanced web metrics with Google Analytics. Hoboken, NJ: John
Wiley & Sons.
Das, R., Turkoglu, I., & Sengur, A. (2009). Effective diagnosis of heart disease through
neural networks ensembles. Expert Systems with Applications, 36(4), 7675–7680.
Davenport, T., & Patil, D. (2012, October). Data Scientist: The Sexiest Job of the 21st Century.
Retrieved from hbr.org/2012/10/data-scientist-the-sexiest-job-of-the-21st-century.
De Mauro, A., Greco, M., Grimaldi, M., & Nobili, G. (2016). Beyond data scientists:
A review of big data skills and job families. Proceedings of IFKAD, 1844–1857.
164 ◾ Analytics and Knowledge Management

De Mauro, A., Greco, M., Grimaldi, M., & Ritala, P. (2017). Human resources for big data
professions: A systematic classification of job roles and required skill sets. Information
Processing & Management, 54(3).
Deetjen, U., Meyer, E.T., & Schroeder, R. (2015). Big data for advancing dementia research:
An evaluation of data sharing practices in research on age-related neurodegenerative
diseases. OECD Digital Economy Papers, No. 246, Paris: OECD Publishing.
Dokoozlian, N. (2016, February). Big data and the productivity challenge for wine grapes.
In Agricultural Outlook Forum 2016 (No. 236854). United States Department of
Agriculture.
Dowell, F. E., Boratynski, T. N., Ykema, R. E., Dowdy, A. K., & Staten, R. T. (2002). Use
of optical sorting to detect wheat kernels infected with Tilletia indica. Plant Disease,
86(9), 1011–1013.
Duhigg, C. (2012). How companies learn your secrets. The New York Times, 16, 2012.
Edwards, S. H., Tilden, D. S., & Allevato, A. (2014, March). Pythy: Improving the intro-
ductory python programming experience. In Proceedings of the 45th ACM technical
symposium on Computer science education (pp. 641–646). New York, NY: ACM.
Fang, W. (2007). Using Google analytics for improving library website content and design:
A case study. Library Philosophy and Practice (e-journal), 121.
Fayyad, U. M., Wierse, A., & Grinstein, G. G. (Eds.). (2002). Information visualization in
data mining and knowledge discovery. San Francisco, CA: Morgan Kaufmann.
Fox, P., & Hendler, J. (2011). Changing the equation on scientific data visualization.
Science, 331(6018), 705–708.
Friendly, M. (2002). Visions and re-visions of Charles Joseph Minard. Journal of Educational
and Behavioral Statistics, 27(1), 31–51.
Gardiner, A., Aasheim, C., Rutner, P., & Williams, S. (2017). Skill requirements in big
data: A content analysis of job advertisements. Journal of Computer Information
Systems, 35, 1–11.
Grossmann, V., Steger, T. M., & Trimborn, T. (2013). The macroeconomics of
TANSTAAFL. Journal of Macroeconomics, 38, 76–85.
Guidi, G., Miniati, R., Mazzola, M., & Iadanza, E. (2016). Case study: IBM Watson ana-
lytics cloud platform as analytics-as-a-service system for heart failure early detection.
Future Internet, 8(3), 32.
Hardin, J., Hoerl, R., Horton, N. J., Nolan, D., Baumer, B., Hall-Holt, O., ... & Ward,
M. D. (2015). Data science in statistics curricula: Preparing students to “think with
data”. The American Statistician, 69(4), 343–353.
Healy, K., & Moody, J. (2014). Data visualization in sociology. Annual Review of Sociology,
40, 105–128.
Hedlund, G. (1994). A model of knowledge management and the N‐form corporation.
Strategic Management Journal, 15(S2), 73–90.
Hey, A. J., Tansley, S., & Tolle, K. M. (2009). The fourth paradigm: Data-intensive scientific
discovery (1st ed.). Redmond, WA: Microsoft Research.
Howe, D., Costanzo, M., Fey, P., Gojobori, T., Hannick, L., Hide, W., ... & Twigger, S.
(2008). Big data: The future of biocuration. Nature, 455(7209), 47–50.
Hoyt, R. E., Snider, D., Thompson, C., & Mantravadi, S. (2016). IBM Watson analytics:
Automating visualization, descriptive, and predictive statistics. JMIR Public Health
and Surveillance, 2(2), e157.
Jarvis, J. (2011). What would Google do?: Reverse-engineering the fastest growing company in
the history of the world. New York, NY: Harper Business.
 Relating Big Data and Data Science to the Wider Concept of KM ◾ 165

Kambaskovic, D., & Wolfe, C. T. (2014). The senses in philosophy and science: From the
nobility of sight to the materialism of touch. A Cultural History of the Senses in the
Renaissance. Bloomsbury, London, 107–125.
Kipping, S., Stuckey, M. I., Hernandez, A., Nguyen, T., & Riahi, S. (2016). A web-based
patient portal for mental health care: Benefits evaluation. Journal of Medical Internet
Research, 18(11), e294.
Kohavi, R. (2001, August). Mining e-commerce data: The good, the bad, and the ugly.
Proceedings of the seventh ACM SIGKDD international conference on Knowledge discov-
ery and data mining (pp. 8–13). San Francisco, CA: ACM.
Lamont, J. (2012). Big data has big implications for knowledge management. KM World,
21(4), 8–11.
Lazer, D., Kennedy, R., King, G., & Vespignani, A. (2014). The parable of Google Flu:
Traps in big data analysis. Science, 343(6176), 1203–1205.
Liebowitz, J. (Ed.). (1999). Knowledge management handbook. Boca Raton, FL: CRC Press.
Lin, X., Wang, P., & Wu, B. (2013, November). Log analysis in cloud computing envi-
ronment with Hadoop and Spark. Broadband network & multimedia technology
(IC-BNMT ), 2013 5th IEEE international conference on (pp. 273–276). IEEE.
Lyon, L., & Mattern, E. (2017). Education for real-world data science roles (Part 2):
A translational approach to curriculum development. International Journal of Digital
Curation, 11(2), 13–26.
Markovikj, D., Gievska, S., Kosinski, M., & Stillwell, D. (2013, June). Mining
Facebook data for predictive personality modeling. Proceedings of the 7th interna-
tional AAAI conference on Weblogs and Social Media (ICWSM 2013) (pp. 23–26).
Boston, MA.
Marz, N., & Warren, J. (2015). Big data: Principles and best practices of scalable realtime data
systems. Shelter Island, New York: Manning Publications.
Matignon, R. (2007). Data mining using SAS enterprise miner (Vol. 638). Hoboken, NJ:
John Wiley & Sons.
McAfee, A., & Brynjolfsson, E. (2012). Big data: The management revolution. Harvard
Business Review, 90(10), 60–68.
Murphy, S. A. (2015). How data visualization supports academic library assessment: Three
examples from the Ohio State University Libraries using Tableau. College & Research
Libraries News, 76(9), 482–486.
Nonaka, I., & Takeuchi, H. (1995). The Knowledge-creating company: How Japanese compa-
nies create the dynamics of innovation. Oxford, UK: Oxford University Press.
O’Leary, D. E. (2017). Emerging white-collar robotics: The case of watson analytics. IEEE
Intelligent Systems, 32(2), 63–67.
Orlikowski, W. J., & Baroudi, J. J. (1991). Studying information technology in organiza-
tions: Research approaches and assumptions. Information Systems Research, 2(1), 1–28.
Padhy, R. P. (2013). Big data processing with Hadoop-MapReduce in cloud systems.
International Journal of Cloud Computing and Services Science, 2(1), 16.
Perkel, J. M. (2015). Pick up python. Nature, 518(7537), 125.
Pollett, S., Boscardin, W. J., Azziz-Baumgartner, E., Tinoco, Y. O., Soto, G., Romero,
C., ... & Rutherford, G. W. (2016). Evaluating Google flu trends in Latin America:
Important lessons for the next phase of digital disease detection. Clinical Infectious
Diseases, 64(1), ciw657.
Provost, F., & Fawcett, T. (2013). Data science and its relationship to big data and data-
driven decision making. Big Data, 1(1), 51–59.
166 ◾ Analytics and Knowledge Management

Raskin, J. (2000). The humane interface: New directions for designing interactive systems.
Boston, MA: Addison-Wesley.
Reber, A. S. (1989). Implicit learning and tacit knowledge. Journal of Experimental
Psychology: General 118(3), 219.
Rittinghouse, J. W., & Ransome, J. F. (2016). Cloud computing: Implementation, manage-
ment, and security. Boca Raton, FL: CRC Press.
Rossini, P. G., Hemsley, J., Tanupabrungsun, S., Zhang, F., Robinson, J., & Stromer-
Galley, J. (July, 2017). Social media, US presidential campaigns, and public opinion
polls: Disentangling effects. Proceedings of the 8th International Conference on Social
Media & Society (p. 56). New York, NY: ACM.
Russom, P. (2011). Big data analytics. TDWI best practices report, fourth quarter, 19, 40.
Sonnenberg, L., Gelsomin, E., Levy, D. E., Riis, J., Barraclough, S., & Thorndike, A. N. (2013).
A traffic light food labeling intervention increases consumer awareness of health and
healthy choices at the point-of-purchase. Preventive Medicine, 57(4), 253–257.
Stark, H., Habib, A., & al Smadi, D. (2016). Network engagement behaviors of three
online diet and exercise programs. Proceedings from the Document Academy, 3(2), 17.
Tableau.com. (2017). www.tableau.com, accessed 2017.
Tal, A., & Wansink, B. (2016). Blinded with science: Trivial graphs and formulas increase
ad persuasiveness and belief in product efficacy. Public Understanding of Science,
25(1), 117–125.
Tang, R., & Sae-Lim, W. (2016). Data science programs in US higher education: An
exploratory content analysis of program description, curriculum structure, and
course focus. Education for Information, 32(3), 269–290.
Telea, A. C. (2014). Data visualization: principles and practice. Boca Raton, FL: CRC Press.
Tenopir, C., Allard, S., Sinha, P., Pollock, D., Newman, J., Dalton, E., ... & Baird, L.
(2016). Data management education from the perspective of science educators.
International Journal of Digital Curation, 11(1), 232–251.
Thompson, G. (2017). Coding comes of age: Coding is gradually making its way from
club to curriculum, thanks largely to the nationwide science, technology, engineering
and mathematics phenomenon embraced by so many american schools. The Journal
(Technological Horizons In Education), 44(1), 28.
Tufte, E. (2001) The visual display of quantitative information (2nd ed.). Cheshire, CT:
Graphics Press.
Tuomi, I. (1999). Corporate knowledge: Theory and practice of intelligent organizations
(pp. 323–326). Helsinki, Finland: Metaxis.
Velardo, S. (2015). The nuances of health literacy, nutrition literacy, and food literacy.
Journal of Nutrition Education and Behavior, 47(4), 385–389.e1.
Vilajosana, I., Llosa, J., Martinez, B., Domingo-Prieto, M., Angles, A., & Vilajosana, X.
(2013). Bootstrapping smart cities through a self-sustainable model based on big data
flows. IEEE Communications Magazine, 51(6), 128–134.
Walker, S. (2014). Big data: A revolution that will transform how we live, work, and think.
International Journal of Advertising, 33(1), 181–183.
Waller, M. A., & Fawcett, S. E. (2013). Data science, predictive analytics, and big data:
A revolution that will transform supply chain design and management. Journal of
Business Logistics, 34(2), 77–84.
Zarcadoolas, C., Vaughon, W. L., Czaja, S. J., Levy, J., & Rockoff, M. L. (2013).
Consumers’ perceptions of patient-accessible electronic medical records. Journal of
Medical Internet Research, 15(8), 284–300.
Chapter 6

Fundamentals of Data
Science for Future
Data Scientists
Jiangping Chen, Brenda Reyes Ayala,
Duha Alsmadi, and Guonan Wang

Contents
Data, Data Types, and Big Data ........................................................................168
Data Science and Data Scientists .......................................................................170
Defining Data Science: Different Perspectives ...............................................170
Most Related Disciplines and Fields for Data Science ...................................172
Data Scientists: The Professions of Doing Data Science ................................173
Data Science and Data Analytics Jobs: An Analysis ........................................... 174
Purposes of Analysis and Research Questions ................................................175
Data Collection ............................................................................................175
Data Cleanup and Integration ...................................................................... 176
Tools for Data Analysis ................................................................................. 176
Results and Discussion..................................................................................177
Characteristics of the Employers...............................................................177
Job Titles ..................................................................................................178
Word Cloud and Clusters on Qualifications and Responsibilities .............178
Summary of the Job Posting Analysis ............................................................182

167
168 ◾ Analytics and Knowledge Management

Data Science Education: Current Data Science Programs and Design

Considerations ..................................................................................................182
Data Science Programs Overview..................................................................182
PhD Programs in Data Science ................................................................182
Masters Programs in Data Science ............................................................186
Graduate Certificate Programs in Data Science ........................................186
Massive Open Online Courses .................................................................186
Bootcamps ............................................................................................... 191
Data Science Program: An Integrated Design ............................................... 191
Summary and Conclusions ................................................................................192
References .........................................................................................................193

Data, Data Types, and Big Data

Data has been frequently discussed along with two other concepts: information and
knowledge. However, the concept of data seems less ambiguous than that of infor-
mation or knowledge. Data are considered to be symbols or raw facts that have not
yet been processed (Ackoff, 1989; Coronel, Morris, & Rob, 2012, p. 5). In contrast,
information has a number of different definitions, such as information as a thing,
something informative, a process, or equivalent to knowledge (Buckland, 1991;
Losee, 1997; Saracevic, 1999; Madden, 2000). Knowledge also does not have an
agreed-upon definition. Davenport and Prusak (1998) viewed knowledge as “a fluid
mix of framed experience, values, contextual information, and expert insight that
provides a framework for evaluating and incorporating new experiences and informa-
tion” (Davenport and Prusak, 1998, p. 5).
The three concepts of data, information, and knowledge are related. They are
products of human intellectual activity at different cognitive levels. Data usually
need to be processed and organized to become informative for human beings, while
knowledge is obtained by making use of data and information through experiences.
There is a fine line between data and information: when we talk about digital infor-
mation or digital data, they are sometimes used interchangeably.
To facilitate our discussion, we follow the idea that data are symbols or raw facts
that are stored in electronic media. They can be produced or collected manually
by humans or automatically by instruments. The following are examples of data in
different forms and formats:

◾ Scientific data: Scientific data are data collected by scientists or scientific

instruments during observations or experiments. Examples of scientific
data include astronomical data collected by telescopes, patients’ vitals
collected by heartbeat monitors, and laboratory data collected by biologists
or physicists.
 Fundamentals of Data Science for Future Data Scientists ◾ 169

◾ Transaction data: Transaction data are data collected by companies to record

their business activities, such as sales records, product specifics, and customer
profiles.
◾ Data that are of public interest: Data that reflects environmental status, human
activities, and social change, such as geographical data and weather forecasts.
◾ Web data: Web data are documents, web pages, images, and videos pub-
lished or released on the Internet. These data also record human activities
and knowledge. They can be raw facts for certain tasks or decision-making
processes.
◾ Social media data: Social media data are generated by Internet users or mobile
users, such as Tweets or postings on social media sites, or comments to others’
postings.

Data can also be classified based on different facets depending on different perspec-
tives. For example, scientific data can be numeric, textual, or graphical. Or, we can
categorize data as structured, semistructured, or unstructured based on whether
they have been organized.
Data type is an important concept in this regard. It has been a basic concept
in computer science. A computer programmer knows that one needs to determine
the data type of a particular variable before applying appropriate operations to
it. Computer languages can usually handle different types of data. For example,
Python, one of the most popular computer programming languages for data analy-
sis, contains simple data types including numerical data type, string, and bytes, and
collection data types such as tuple, list, set, and dictionary. Data might need to be
changed to a different type to be processed properly.
Most data we talk about in this chapter are actually digital data in elec-
tronic form. Data are generated every day and at every moment. One of the
characteristics of data is that it can be copied easily and precisely, or transferred
from one medium to another with great speed and accuracy. However, that
is not always true, especially when we start to deal with Big Data. Big Data
refer to digital data with three Versus: (1) volume, (2) variety, and (3) velocity
(Laney, 2011). Volume refers to the situation that data grows at a rapid rate,
variety refers to the many data types in which data exist, and velocity refers to
the speed of data delivery and processing. Dealing with Big Data requires theo-
ries, methods, technologies, and tools, which lead to the emergence of data sci-
ence, a new discipline that focuses on data processing, which will be discussed
in the remaining sections.
Data are an important resource for organizations and individuals who use them
to make decisions. When Big Data becomes publicly available, no one can ignore
the huge potential impact of Big Data on business, education, and personal lives.
Therefore, data science as a new discipline has drawn great attention from govern-
ments, industry, and educational institutions.
170 ◾ Analytics and Knowledge Management

Data Science and Data Scientists

Dr. William S. Cleveland formally defined data science in 2001 as it is used today
in his article “Data Science: An Action Plan for Expanding the Technical Areas
of the Field of Statistics” (Cleveland, 2001). Since then, the concept of data sci-
ence has been further developed and has been increasingly linked to data analytics
and Big Data. Companies have realized the value of data science and its useful-
ness in discovering knowledge and helping with decision-making. The need for
data science workers, or data scientists, has increased tremendously in recent years
(Leopold, 2017).

Defining Data Science: Different Perspectives

Providing a precise definition for a discipline is important, especially for students
who are eager to acquire important knowledge and skills for future job seeking.
This section summarizes the definitions of data science and analyzes the features
of this discipline. We also provide our own definition based on our understanding
of this important field.
The following are prominent data science definitions we found in the literature:

◾ Data science as an expansion of the technical areas of the field of statistics.

(Cleveland, 2001)
◾ “Data science involves principles, processes, and techniques for understanding
phenomena via the (automated) analysis of data.” (Provost & Fawcett, 2013)
◾ “Data science is the study of the generalizable extraction of knowledge from
data.” (Dhar, 2013)
◾ “An extension of information systems research.” (Agarwal & Dhar, 2014)
◾ “This coupling of scientific discovery and practice involves the collection,
management, processing, analysis, visualization, and interpretation of vast
amounts of heterogeneous data associated with a diverse array of scientific,
translational, and interdisciplinary applications.” (Donoho, 2015)
◾ “Data science is now widely accepted as the fourth mode of scientific discov-
ery, on par with theory, physical experimentation and computational analy-
sis. Techniques based on Big Data are showing promise not only in scientific
research, but also in education, health, policy, and business.” (Michigan
Institute for Data Science, 2017)
◾ “Data science enables the creation of data products.” (Loukides, 2011, p. 1)
◾ “Data science is the study of where information comes from, what it repre-
sents, and how it can be turned into a valuable resource in the creation of
business and IT strategies.” (Banafa, 2014)
◾ “Data science is not only a synthetic concept to unify statistics, data analysis
and their related methods but also comprises its results. It includes three phases,
design for data, collection of data, and analysis on data.” (Hayashi, 1998, p. 41)
 Fundamentals of Data Science for Future Data Scientists ◾ 171

◾ “Data science is a combination of statistics, computer science, and informa-

tion design.” (Shum et al., 2013)
◾ “Data science is a new trans-disciplinary field that builds on and synthesizes
a number of relevant disciplines and bodies of knowledge, including statis-
tics, informatics, computing, communication, management, and sociology,
to study data following data thinking.” (Cao, 2017)

The above definitions reflect current understanding of this discipline. Together

they help to describe the characteristics of data science, which can be summarized
into the following aspects:

◾ The center of data science is data. Especially, Big Data becomes the subject
that is investigated.
◾ The purpose of data science is to obtain information or knowledge from data.
The information will help to make better decisions, and the knowledge may
help an organization, a state, a country, or the whole of humanity to better
understand the development or change of the nature or the society.
◾ Data science is a multidisciplinary field that has applied theories and
technologies from a number of disciplines and areas such as mathemat-
ics, statistics, computer sciences, information systems, and information
science. It is expected that data science will bring change to impact these
disciplines.

We define data science as an interdisciplinary field that explores scientific meth-

odology and computational technology about data, including data management,
access, analysis, and evaluation for the benefits of human beings.
In our definition, data are at the center of data science. Any activities around
data should be considered within the scope of this field. In other words, data sci-
ence not only deals with data mining or data analysis, but also explores technolo-
gies and approaches for other important activities including data management, data
access, and evaluation. We avoid listing related disciplines in the definition because
we believe data science has impact and promises new opportunities to every scien-
tific and industry field.
As an emerging discipline, data science is still in its evolving stage. More
research and investigation are needed to understand data science problems,
challenges, theories, and methodologies. Interesting questions in data science
to explore can be: what are its other major concepts in addition to data? What
are the problems data science should attempt to tackle? And do we have enough
theories and frameworks to guide future practice in data science? As specified by
Provost & Fawcett (2013), in order for data science to flourish as a separate field,
we should think beyond the specific algorithms, techniques, and tools that we
are currently using and work on developing the core principles and concepts that
underlie related techniques.
172 ◾ Analytics and Knowledge Management

Most Related Disciplines and Fields for Data Science

Data science has been considered an interdisciplinary field since its emergence. It
was originally developed within the statistics and mathematics community (Cao,
2017). At its current stage, data science is more closely related to several disciplines
than others, which is evident from the definitions in the literature provided in the
previous section. Even though we believe data science is relevant to every discipline,
some disciplines or fields have more impact on data science than others. These
disciplines include mathematics (especially probability and statistics), information
science, computer science, knowledge management, management information sys-
tems, and decision science. The data science programs at the doctoral and master’s
levels (see Tables 6.3 and 6.4) confirm the above list, as most of the data science
programs were established in colleges, schools, or departments of information sci-
ence, computer science, statistics, and business management.
Many studies have explored the connections between data science and math-
ematics, computer science, and management information systems (Cleveland,
2001; Shum et al., 2013; Agarwal & Dhar, 2014). Mathematics including statistics
provides numerical reasoning and methodologies for data analysis and processing.
Some mathematical concepts and theorems such as those in calculus, linear alge-
bra, and statistics needed to be mastered by potential data science workers; com-
puter science provides programming languages, database systems, and algorithms
that a data science worker needs to manipulate numerical and textual data; man-
agement information systems, on the other hand, have explored data mining and
data analytics for years.
However, we found that the connection of data science with information sci-
ence was largely neglected which may negatively impact the development of data
science as a discipline. The impression of the public on information science is
actually library science, which has focused more on classic information resource
management, library management, and reference services in libraries. Even though
information science has been closely related to library science in history, its land-
scape has dramatically changed since the advent of the Internet. Information sci-
ence can definitely provide useful theories and guidance to the development of
data science. For example, one of the important concepts to explore in data science
may be the data lifecycle. There are many ways to describe and depict data lifecycle
(Chang, 2012). For data management, the data lifecycle can include creating data,
processing data, analyzing data, presenting data, giving access to data, and reus-
ing data (Boston University Libraries, n.d.). Or it can be a series of processes to
create, store, use, share, archive, and destroy data (Spirion, n.d.). Explanations of
the different stages included in the lifecycle can be different. But some processes,
such as archiving, storing, and accessing have been well explored for information
in information science.
Data science is also closely related to knowledge management. Hawamdeh
(2003) describes the field of knowledge management and its relationship with
 Fundamentals of Data Science for Future Data Scientists ◾ 173

Value

Step 5: Use
of information
and knowledge
Step 4: Information
management and Knowledge
knowledge sharing management
Step 3: Data/information Data science
processing, analysis, and presentation
Information science
Step 2: Data collection, Information
acquisition and content management

Step 1: Information systems and

infrastructure
Investment

Figure 6.1 Data science in the context of knowledge process. (Adapted from
Hawamdeh, S., Knowledge Management: Cultivating Knowledge Professionals,
Chandos Publishing, Oxford, UK, 2003.)

other disciplines for the purpose of educating knowledge professionals (p. 168).
He considered knowledge management a multidisciplinary subject. Knowledge
management professionals need to acquire information technology and informa-
tion science skills, which serve as foundation for higher-level knowledge work such
as knowledge management and sharing. We expand on his framework by includ-
ing an additional step, shown in Figure 6.1, called data processing, analysis, and
processing, between the original step 2 (information acquisition and content man-
agement) and step 3 (information and knowledge sharing). In this consideration,
information science, data science, and knowledge management are closely related
and overlapping disciplines or areas. Typically, data science centers on activities in
step 2, but should also include activities and tasks in steps 2, 4, and 5.

Data Scientists: The Professions of Doing Data Science

The increasing interest in data science calls for more data science workers. Several
reports indicated that data scientists, which is very frequently called the “sexiest
job of the twenty-first century” (Davenport & Patil, 2012), are in high demand.
McKinsey & Company (2011) pointed to a fact that the United States is facing a
shortage of 140,000–190,000 analytical and managerial skills necessary to handle
business processes related to critical decision-making and Big Data. Additionally,
companies in other sectors have started recruiting data scientists as a crucial
174 ◾ Analytics and Knowledge Management

component to conduct their business strategy. For example, Greylock Partners, a

leading venture capital firm in Silicon Valley that was an early investor in such well-
known companies such as Facebook and LinkedIn, is now showing more concern
about expanding its team-based portfolio and recruiting more talented people in
data science and analytics fields (Davenport & Patil, 2012).
A data scientist is usually described as “a hybrid of data hacker, analyst, com-
municator, and trusted adviser” (Davenport & Patil, 2012, p. 73). Data scien-
tists’ jobs “involve data collecting, cleaning, formatting, building statistical and
mathematical models, and communicate professionally with their partners and
clients” (Carter & Sholler, 2016). According to Stodder (2015), data scientists
begin by analyzing business problems, and then providing strategic and mean-
ingful insights about probable actions and future plans. This strategy involves
identifying business drivers, establishing a creative team which is characterized
by excellent technical and communication skills, and using the latest analytics
and visualization techniques (Stodder, 2015). Provost and Fawcett (2013) specified
that successful data scientists “must be able to view business problems from a data
perspective.”
The literature also describes knowledge and skills that are necessary for data
scientists. For example, Dhar (2013) believed that “a data scientist requires an inte-
grated skill set spanning mathematics, machine learning, artificial intelligence, sta-
tistics, databases, and optimization, along with a deep understanding of the craft of
problem formulation to engineer effective solutions.” He considered four basic types
of courses that contribute to the knowledge and skills for a qualified data scientist:
machine learning, computer science (data structure, algorithms, and systems), cor-
relation and causation, and problem formulation.
Data scientist can be a general term for people working in data science. Different
titles can be found from job postings and literature for related positions. It is important
to realize that data scientists may vary in their background, experiences, skills, and ana-
lytical workflows. In semistructured interviews of 35 data analysts, Kandel, Paepcke,
Hellerstein, & Heer (2012) indicated that analysts fall into three categories which are
hacker, scripter, and application user, each with different levels of technical skills and
tasks that can be performed. They, however, generally share five common tasks that sup-
port daily insights, which are discovery, wrangling, profiling, modeling, and reporting.
To verify what we have learned from the literature and to understand the
increase in the demand for data scientists (Leopold, 2017), we conducted a small-
scale job analysis. The analysis is reported in the next section.

Data Science and Data Analytics Jobs: An Analysis

One of the goals of this chapter is to explore how higher education could produce
qualified data science workers, or data scientists. We therefore conducted a pre-
liminary job analysis to understand current job markets for data science workers.
 Fundamentals of Data Science for Future Data Scientists ◾ 175

This section will present the purposes of the analysis, the research questions, the
process of data collection and analysis, and the results.

Purposes of Analysis and Research Questions

This analysis aimed to understand the current job market for data scientists. Such
understanding would provide us with guidelines to designing data science courses
and programs at undergraduate and graduate levels. Specifically, we would like to
find answers to the following questions:

1. What characterizes the employers who hire data scientists?

2. What were the job titles used when employers are looking for qualified data
science workers?
3. What are the general qualifications, knowledge, tools, or skills required by
employers?

As mentioned earlier, our analysis also served as a validation of knowledge, skills,

and competences for qualified data scientists addressed in the literature.

Data Collection
We recruited the students of a graduate-level information science class at the
University of North Texas to collect the data. INFO 5717 is a course in which
students learn to build web-based database systems. The students were assigned to
five teams with three members in each team. They were asked to collect, process,
and upload at least 60 job postings pertaining to data analysts, information man-
agement specialists, data scientists, and business intelligence analysts to a MySQL
database.
Prior to data collection, we provided a list of metadata elements to guide the
process. Students were required to collect information for at least the following
elements for each job posting:

◾ Title of the job

◾ Start date and end date for the position
◾ Name of the employer
◾ Salary
◾ URL of the job posting at the employer’s website
◾ Position requirements
◾ Position responsibilities

Students were instructed to collect the above data on the Internet using methods
they considered appropriate. The data was collected from job aggregator websites
176 ◾ Analytics and Knowledge Management

such as Linkedin.com, indeed.com, dice.com, and glassdoor.com in November

2016. Additionally, some of the teams also collected data pertaining to the employ-
ers such as the URL of the official employer website, contact email, and the mailing
address. The data was initially saved as text files.
We also used this project as an opportunity to train students for future data
science-related skills. Therefore we asked students to perform a series of tasks to
process the collected data: each student team was required to design a job database
using MySQL, containing one or two database tables to store the job posting and
the employer information. Using HTML and PHP scripts, students created simple
web entry forms to insert the job and employer information into their databases. As
part of their course requirements, students developed web pages to manipulate or
send queries to retrieve information from the data. They also developed authentica-
tion and session-control mechanisms to secure and use their database. In total, the
teams collected 364 job postings.

Data Cleanup and Integration

After the students completed the classwork, we exported the job and employer
data from five databases to Excel spreadsheets for analysis. We found that even
though we gave students metadata specifications, the database tables from each
team were not completely aligned—some had more elements, or the same elements
were called different names. Employer information was more problematic because
we did not specify what exactly information should be collected about employers
except that we needed to have the name and URL of the employer for each job
posting. Furthermore, students might have collected the same job posting (dupli-
cation) as the teams worked independently of each other. Some of the collected
records had some missing values such as job location and salaries because the infor-
mation pertaining to the corresponding job was not stated clearly in the original
announcements.
We conducted manual data cleanup to fix the above data issues. Some minor
cases of missing salary values were handled by using online popular salaries esti-
mate tools such as the job search engine indeed.com (Doherty, 2010). The manual
data cleanup was possible because we had only 364 job postings. After removing
the duplicates and consolidating the data elements, we obtained 298 usable obser-
vations for analysis.

Tools for Data Analysis

For this small dataset, we mainly used Excel to sort the different elements, to gen-
erate frequency counts and percentages, and to produce diagrams. Additionally,
SAS Enterprise Miner (SASEM) 14.1 was utilized to perform cluster analysis (SAS
Institute, 2014) on qualifications and required experience as presented in the job
 Fundamentals of Data Science for Future Data Scientists ◾ 177

postings. We also wrote a simple program using the R language to extract the most
frequent keywords and produced a visual representation for the results called a word
cloud (Heimerl, Lohmann, Lange, & Ertl, 2014). Manual data transformation and
adjustment were performed to use these tools for the purpose of answering the
research questions proposed in Section 3.1.

Results and Discussion

In this section, we present the results of our analysis and answer our research
questions.

Characteristics of the Employers

From the 298 valid job postings, 236 company names were identified by the stu-
dents. Among them we found some well-known companies such as Apple, Blue
Cross and Blue Shield, Capital One, Bank of America, Ericsson, Google, IBM,
Infosys, Nike, Samsung, Sprint, Verizon, Walmart, Wells Fargo, and Xerox. These
employers covered different industrial fields such as information technology,
healthcare, insurance, finance, retail, biotechnology, business services, manufac-
turing, and media sectors.
With respect to job postings’ geographical concentration, Texas leads the United
States with more than 24% of the jobs posting, and next comes Illinois and California
with 17% and 15% respectively (Figure 6.2). The collected data also shows that both
29
21
18

11
9
8

Total
5
4

3
1

1
1

1
1

1
1
1
1
A a
lif ta
lo ia
Fl do
G ida
Ill gia

n n
N isso ta

nn ina

Te ee
W Vir as
hi ia
riz s

K ois

M chi s

or w ey

Te rol k

n
N Ne Jers i
e
n

i a

r
in ga

Ca or

to
Co orn

as gin
Ca tlan

M eso

x
at

s
ew u
M ans
o

ra

r
or

in

es
th Y

ng
eo
St
A
d
te
ni
,U
re
he
w
ny
A

Figure 6.2 Job distributions among states.

178 ◾ Analytics and Knowledge Management

graduates and undergraduates were possible candidates for data science jobs. The
preferred levels of experience ranged from two to more than 10 years. These results
portrayed an image of the job market for data science.

Job Titles
A frequency analysis of the 298 job postings showed that the top job titles used
by the employers were data scientist, data analyst, business intelligence analyst
or intelligence analyst, information management specialist or analyst, and data
engineer. For data scientist and data analysts, the job title could specify different
levels, such as lead or principal, senior, junior, or intern. Some job titles, such as
data analytics scientist and data science quantitative analyst, were a mixture of data
analyst, data science, or data scientist. Other titles included Big Data architect, data
integrity specialist, data informatics scientist, marketing analyst, MIS specialist,
project delivery specialist, and consultant on data analytics (Table 6.1).

Word Cloud and Clusters on Qualifications and Responsibilities

We used the R programming language to visually summarize the most frequent
terms that occurred in two fields: qualifications and responsibilities. The word cloud
in Figure 6.3 shows that terms like “experience,” “business,” “analyst,” “analysis,”
“management,” and “skills” stood out for their frequent occurrences, as depicted in
Figure 6.1. This was supported by many field researchers, career practitioners, and
job posters who confirmed that effective data scientists and analysts need to have a
substantial level of managerial and business experience in addition to sophisticated
skills in programming, development, design, and statistics (Davenport & Patil,
2012; Kandel et al., 2012).

Table 6.1 Job Titles

Job Titles Frequency

Data Scientist 95 (31.9%)

Data Analyst 95 (31.9%)

Business Intelligence Analyst (Intelligence Analyst) 50 (16.8%)

Information Management Specialist/Analyst 13 (4.4%)

Data Engineer 10 (3.3%)

Research Analyst 5 (1.7%)

Other 30 (10%)

Total 298
 Fundamentals of Data Science for Future Data Scientists ◾ 179

Figure 6.3 Most frequent terms in qualifications and responsibilities.

We also conducted cluster analysis on these two fields. The purpose of the cluster
analysis was to identify concepts that are implied in the dataset. The process starts
by exploring the textual data content, using the results to group the data into mean-
ingful clusters, and reporting the essential concepts found in these clusters (SAS
Institute, 2014). Using the SASEM cluster analysis function, which is based on
the mutual information weighting method, we obtained five clusters as presented
in Table 6.2. According to the terms in each cluster, we could name cluster 1 as
project management; cluster 2 as machine learning and algorithmic skills; cluster 3
as statistical models and business analytics; cluster 4 as database management and
systems support, and cluster 5 as communication skills. Among them, cluster 2 is
the largest cluster with a relative frequency of 47% among all the terms in these
two fields.
The text filter node functionality in SAS enabled us to invoke a filter viewer to
explore interactive relationships among terms (SAS Institute, 2014). To preview the
180 ◾ Analytics and Knowledge Management

Table 6.2 The Clusters of the Job Postings

Frequency
Cluster ID Cluster Name Descriptive Terms (Percentage)

1 Project Stakeholder, project, 46 (15%)

management presentation,
recommendation,
communicate, plan,
understand, fast-pace,
requirement, manage, team
multiple, issue, and
improvement

2 Machine learning Statistical, technique, 144 (47%)

and algorithmic machine, model, test,
skills learning, system, report,
create, design, requirement
multiple, and implement

3 Statistic models Sale, market, customer 60 (19%)

and business analytics, help analytic,
analytics build, insight, service,
opportunity, engineer key,
solution, and team

4 Database Request, assign, perform, 37 (12%)

management and process, procedure
systems support information, maintain,
database, improvement,
relate, assist, support, issue,
and service

5 Communication Ability, experience, skill 21 (7%)

skills strong, fast-pace highly,
demonstrate,
communication, write
multiple, service analytical,
and customer information

concepts that are highly correlated with the term data, a concepts link map was
generated, as depicted in Figure 6.4. The five clusters in Table 6.2 become the six
nodes (two notes on analysis/analyze) with links to other outer notes.
The word cloud and cluster analysis showed a broader view of the terms from the
job market perspective. Also, further exploration of the concepts link map revealed
that data scientists need to have substantial skills in Hadoop, Hive, Pig, Python,
Machine Learning, C++ and Java programming, R, and SQL, in addition to being
proficient with statistical and modeling packages such as SPSS and SASEM.
 + sas institute + result effectively + tool
actionable
business performance + member
+ methodology + data scientist
+ present + process

business + scientist
+ communicate + project
+ pattern + team
+ work

+ interpret
+ organization

+ source
+ market
+ analyze
performance
metrics

data + insight
+ question
+ business insight

+ actionable insight
+ product

+ opportunity

+ practice
statistical
+ analysis + recommendation
+ model
+ algorithm
+ presentation
predictive
multiple + trend
+ test
+ perform
+ build + conduct
+ data model + perform
+ design
+ predictive model data analysis
Fundamentals of Data Science for Future Data Scientists
◾

Figure 6.4 The concepts link map of the job postings.

181
182 ◾ Analytics and Knowledge Management

Summary of the Job Posting Analysis

This small-scale study covers many activities a data analyst may perform according
to Donoho (2015). We realized that a data science worker needs to know how to
organize and clean up data, how to analyze data, and how to create data visualiza-
tions using different tools. It would be very beneficial if the data analyst also pos-
sesses basic research skills, such as how to collect data and how to evaluate it. Data
analysis also involves team-based work of data scientists who use effective tools to
build statistical and predictive models to produce business and actionable insights
within their organizations, which should be also built in data science curriculum.
Our job analysis above provides insights into this new data science discipline
and its career opportunities. These insights can guide educators to develop appro-
priate data science programs and courses that meet the short-term and long-term
needs of the digital economy. Next, we present our thoughts on data science educa-
tion based on this job analysis and an overview of existing data science programs.

Data Science Education: Current Data Science

Programs and Design Considerations
The increasing demand for data science workers has motivated the rapid growth of
data science-related programs, as presented in this section. Universities and colleges
have been actively developing new data science programs and courses at different
levels to meet the needs of the job market on qualified data science workers. It is
necessary to review and summarize what has been established so far to develop
high-quality data science programs.

Data Science Programs Overview

As we have discussed earlier, data science is an interdisciplinary field. It integrates
and applies knowledge and techniques in multiple disciplines and fields. Data sci-
ence programs and courses have been established to meet the needs for training
effective data scientists. These programs offer a variety of courses to train students
with the research and professional skills to succeed in leading edge organizations.
This section summarizes existing data science programs and trainings.

PhD Programs in Data Science

According to a data science salary study, 88% of data scientists have advanced
degrees and 46% have a PhD (Burtch Works Executive Recruiting, 2017). As of
June 2017, the website MastersInDataScience.org lists 18 universities offering
doctorate-level programs in data science and data analytics, as showed in Table 6.3.
The majority of these PhD programs require students to have undertaken tech-
nical coursework or earned a technical degree in computer science, mathematics,
Table 6.3 PhD Programs in Data Science
University School or Department Degree, Online or Campus, and URL

Colorado Technical University Computer Science Department Doctor of Computer Science with a
concentration in Big Data Analytics, online,
http://www.coloradotech.edu/degrees/
doctorates/computer-science/
big-data-analytics

Indiana University Bloomington School of Informatics and Computing PhD minor in Data Science, campus or online,
http://www.soic.indiana.edu/graduate/degrees/
data-science-minor.html

Brown University Computer Science Department PhD in Computer Science—concentration in

Data Science, campus, http://cs.brown.
edu/~kraskat/phd.html

Chapman University Schmid College of Science and PhD in Computational and Data Sciences,
Technology campus, http://www.chapman.edu/scst/
graduate/phd-computational-science.aspx

Georgia State University Department of Computer Science PhD (Bioinformatics Concentration), campus,
http://cs.gsu.edu/graduate/doctor-philosophy/
ph-d-bioinformatics-concentration-degree-
requirements/
Fundamentals of Data Science for Future Data Scientists

Indiana University-Purdue School of Informatics and Computing PhD in Data Science, campus, https://soic.iupui.
◾

University-Indianapolis edu/hcc/graduate/data-science-phd/

(Continued)
183
Table 6.3 (Continued) PhD Programs in Data Science
184 ◾

University School or Department Degree, Online or Campus, and URL

Kennesaw State University Department of Statistics and Analytical PhD in Analytics and Data Science, campus,
Sciences http://csm.kennesaw.edu/datascience/

New York University Tandon School of Engineering PhD in Computer Science with Specialization in
Visualization, Databases, and Big Data,
campus, http://engineering.nyu.edu/
academics/programs/computer-science-phd

Newcastle University School of Computer Science EPSRC CDT in Cloud Computing for Big Data,
campus, http://www.bigdata-cdt.ac.uk/

Oregon Health and Science Department of Medical Informatics and PhD in Biomedical Informatics—Bioinformatics
University Clinical Epidemiology and Computational Biology Track, Clinical
Informatics Track, campus, http://www.ohsu.
edu/xd/education/schools/school-of-medicine/
departments/clinical-departments/dmice/
Analytics and Knowledge Management

educational-programs/clinical-informatics.cfm

University of Cincinnati College of Medicine PhD in Biostatistics—Big Data Track, campus,

http://med.uc.edu/eh/divisions/bio/programs/
phd-biostatistics-big-data-track

University of Maryland-College College of Information Studies PhD in Information Studies—Concentration in

Park Big Data/Data Science, campus, https://ischool.
umd.edu/phd-admissions

(Continued)
Table 6.3 (Continued) PhD Programs in Data Science
University School or Department Degree, Online or Campus, and URL

University of College of Management PhD in Business Administration—information

Massachusetts-Boston systems for data science track, campus,
https://www.umb.edu/academics/cm/
business_administration_phd/mis

University of Michigan-Ann Department of Statistics PhD in Statistics, campus, http://lsa.umich.edu/

Arbor stats/graduate-students/graduate-degree-
programs/phdprograms.html

University of Pittsburgh Graduate School of Public PhD in biostatistics, campus, http://www.

publichealth.pitt.edu/biostatistics/academics/
phd

University of Southern Marshall School of Business PhD in Data Sciences and Operations, campus,
California https://www.marshall.usc.edu/index.php/
departments/data-sciences-and-operations

University of eScience Institute PhD in Big Data and Data Science, campus,
Washington-Seattle http://escience.washington.edu/education/phd/
igert-data-science-phd-program/

Worcester Polytechnic Institute College of Arts and Sciences PhD in Data Sciences, campus, https://www.wpi.
Fundamentals of Data Science for Future Data Scientists

edu/academics/study/data-science-phd
◾
185
186 ◾ Analytics and Knowledge Management

or statistics as a prerequisite for entrance. These programs are on-campus and

require students to be present in class to complete their degrees; only a couple of
programs offer online classes but still have the requirement of attending campus
activities to obtain the degree.

Masters Programs in Data Science

The same data science salary study also mentioned that 59% of data scientists
who are early in their career hold a master’s degree, a significant increase from
48% in 2015 (Burtch Works Executive Recruiting, 2017). As of June 2017, the
site MastersInDataScience.org lists 23 schools offering master’s degree programs in
data science, as showed in Table 6.4.
These master’s degree programs have a duration from nine months to about two
years, depending on curriculum requirements and the part-time or full-time status
of the students. Students usually begin by taking required courses specified in the
curriculum in the first semester. Once they obtain basic knowledge and become
familiar with technologies in data science, such as programming, databases, and
statistics, they proceed to do practicums, application, and capstone projects. Some
of the schools provide internship opportunities and practicums so students can work
with partner companies in industry or government agencies during their study.

Graduate Certificate Programs in Data Science

Certificate programs provide a flexible way for people to study data science. A certifi-
cate program generally takes less time and costs less money than a program intended
for earning a degree. It is an attractive option for people who are interested in pursu-
ing a career in this field. The website MastersInDataScience.org has a list of 92 uni-
versities offering graduate certificate programs in data science and related specialties.

Massive Open Online Courses

Massive open online courses (MOOCs) are online courses with open (mostly free)
access via the Internet. These courses are similar to university courses, but do not
tend to offer academic credit. They allow students to choose their own academic
path and complete the courses on their own schedule. To study data science or data
analysis, students may choose courses from MOOCs, such as Coursera, and con-
struct a learning plan primarily from the following sample areas:

◾ Programming languages: The most popular languages currently used are

Python and R for data science courses.
◾ Statistics and probability: There are broad theories that are used for making
informed choices in analyzing data and for drawing conclusions about scien-
tific truths from data.
Table 6.4 Master Programs in Data Science
University School/College Degree, Online Option, URL

Southern Methodist University Dedman College of Humanities and Master of Science in Data Science,
Sciences, Lyle School of Engineering online option, https://requestinfo.
and Meadows School of the Arts datascience.smu.edu/index10.html

University of California, Berkeley School of Information Master of Information and Data Science,
online option, https://requestinfo.
datascience.berkeley.edu/index3.html

Arizona State University W.P. Carey School of Business Master of Science in Business Analytics,
online option, https://programs.
wpcarey.asu.edu/masters-programs/
business-analytics

Carnegie Mellon University School of Computer Science Master of Computational Data Science,
no online option, https://mcds.cs.cmu.
edu/

Columbia University Data Science Institute Master of Science in Data Science, no

online option, http://datascience.
columbia.edu/master-of-science-in-
data-science

Cornell University Department of Statistical Science Master of Professional Studies in

Applied Statistics (Option II: Data
Fundamentals of Data Science for Future Data Scientists

Science), no online option, http://stat.

◾

cornell.edu/academics/mps

(Continued)
187
Table 6.4 (Continued) Master Programs in Data Science
188 ◾

University School/College Degree, Online Option, URL

Georgia Tech College of Computing, College of Master of Science in Analytics, no

Engineering, and Scheller College of online option, https://analytics.gatech.
Business edu/

Illinois Institute of Technology College of Science Master of Data Science, online option,
http://science.iit.edu/programs/
graduate/master-data-science

Indiana University, Bloomington School of Informatics and Computing Master of Science in Data Science,
online option, http://www.soic.indiana.
edu/graduate/degrees/data-science/
index.html

New York University Center for Data Science Master of Science in Data Science, no
online option, http://cds.nyu.edu/
academics/ms-in-data-science/
Analytics and Knowledge Management

North Carolina State University Institute for Advanced Analytics Master of Science in Analytics, no
online option, http://analytics.ncsu.
edu/

Northwestern University McCormick School of Engineering and Master of Science in Analytics, and
Applied Science, and School of Master of Science in Predictive
Continuing Studies Analytics, http://www.mccormick.
northwestern.edu/analytics/

(Continued)
Table 6.4 (Continued) Master Programs in Data Science
University School/College Degree, Online Option, URL

Rutgers, The State University of New Computer Science Department Master of Science in Data Sciences, no
Jersey online option, https://msds-cs.rutgers.
edu/msds/aboutpage

Stanford University Department of Statistics Master of Science in Statistics: Data

Science, no online option, https://
statistics.stanford.edu/academics/
ms-statistics-data-science

Texas A & M University Department of Statistics Master of Science in Analytics, online

option, http://analytics.stat.tamu.edu/

University of California, San Diego Departments of Computer Science and Master of Advanced Study in Data
Engineering Science and Engineering, no online
option, http://jacobsschool.ucsd.edu/
mas/dse/

University of Illinois at Department of Computer Science, Professional Master of Computer

Urbana-Champaign Department of Statistics, School of Science in Data Science, online option,
Information Sciences and Coursera https://online.illinois.edu/mcs-ds

University of Minnesota—Twin Cities College of Science and Engineering, Master of Science in Data Science, no
College of Liberal Arts and School of online option, https://datascience.umn.
Fundamentals of Data Science for Future Data Scientists

Public Health edu/

(Continued)
◾ 189
Table 6.4 (Continued) Master Programs in Data Science
190 ◾

University School/College Degree, Online Option, URL

University of San Francisco College of Arts and Sciences Master of Science in Analytics, no
online option, https://www.usfca.edu/
arts-sciences/graduate-programs/
analytics

University of Southern California Viterbi School of Engineering Master of Science in Computer

Science—Data Science, online option,
https://gapp.usc.edu/graduate-
programs/masters/computer-science/
data-science

University of Virginia Data Science Institute Master of Science in Data Science, no

online option, https://dsi.virginia.edu/
academics

University of Washington—Seattle Interdisciplinary Master of Science in Data Science, no

Analytics and Knowledge Management

Campus online option, https://www.

datasciencemasters.uw.edu/

University of Wisconsin Department of Statistics, and UW Master of Science in Statistics—Data

Extension Science, and Online Master of Science
in Data Science, https://
advanceyourcareer.wisc.edu/degrees/
data-science/
Source: http://www.mastersindatascience.org.
 Fundamentals of Data Science for Future Data Scientists ◾ 191

◾ Data collection: Extracting data from the web, from APIs, from databases,
and from other sources.
◾ Data cleaning and managing: Manipulating and organizing the data collected
to make them useful for data analysis tasks.
◾ Exploratory data analysis: Exploring data to understand the data’s underlying
structure and summarizing the important characteristics of a dataset.
◾ Data visualization: Using plotting systems to construct data graphics, analyz-
ing data in graphical format, and reporting data analysis results.
◾ Machine learning: Building prediction functions and models by using super-
vised learning, unsupervised learning, and reinforcement learning.

Bootcamps
Bootcamp programs are non-traditional educational paths. Compared with tradi-
tional degrees, these programs are intense and have faster routes to the workplace.
It is another education option for considering a career as a data scientist. Data science
bootcamps provided by DataScience.Community (Datascience.Community, n.d.)
has a list of bootcamps available for data science and data engineering.
In summary, data science programs at undergraduate, master, and doctoral level
institutions have been developed in the United States. These programs provide a
wide range of choices to students who want to obtain knowledge and skills in data
science. Still, more data science programs are being developed. It may be the time
to explore the characteristics of a competitive data science program.

Data Science Program: An Integrated Design

Based on our understanding of data science and its related concepts, the knowledge
and skills required for data science workers, and current data science programs, we
believe that a high-quality data science program should provide courses and train-
ing in the following areas:

◾ Fundamental concepts, disciplines, and the profession: The program should offer
one or two courses to introduce the student to basic concepts, disciplines,
and the profession of data science. These courses set up a solid foundation for
students to learn more advanced concepts and applications. They may teach
not only concepts and characteristics of data, information, knowledge, and
the data lifecycle, but also related mathematical concepts, functions, models,
and theorems. Students should develop an affection for data and be willing
to do data science.
◾ Statistical analysis and research methodology: The program should offer multi-
ple courses to teach data collection and data analysis skills, which are usually
taught in master or doctoral level research methodology classes.
192 ◾ Analytics and Knowledge Management

◾ Programming languages, algorithms, and tools: Courses such as computer algo-

rithms, programming languages, data mining, database design, and machine
learning help students to apply computational techniques for data collecting,
cleaning, analysis, prediction, and presentation.
◾ Business logic and soft skills: Students need to understand the purpose of data
science and they need to learn how to communicate and collaborate with dif-
ferent business units such as sales, marketing, and production. Also, the need
to learn to effectively present data analysis results is also important.
◾ Practicum or a capstone project: Data science students should do at least one
project that integrates what they have learned to address a challenging real-
world data science problem. The ideal case would be to have the student work
in a company or an organization to participate in real data science projects
that may help the organization to make better business decisions.

Each student can choose different courses or focus on different areas to improve
their knowledge and skills based on their backgrounds and personal interests. For
example, a student with a business background may want to take mainly courses
on data management and programming languages, while, an information science
student should study business logic in addition to statistical analysis and data visu-
alization through coursework or the practicum.
A course may also teach a student knowledge and skills in different areas. For
example, a database course may teach students data collection, cleaning, analysis,
and report writing through its term project. The instructor can also design projects
consisting of research values or reflecting real industry needs.
In general, a flexible curriculum and close connections with profit and non-
profit organizations provide students with much flexibility and opportunities to
become successful data workers.

Summary and Conclusions

This chapter explored the concept and characteristics of data. It reviewed dif-
ferent perspectives from the literature on data science and provided the authors’
definition of this emerging field. The interdisciplinary nature of data science was
discussed. We then presented a preliminary analysis of about 300 job postings on
data scientists and related positions. Based on the literature review and the job
posting analysis, we summarized the knowledge and skills required for different
levels of data science workers. Furthermore, we reviewed current data science
programs, focusing on the shared values and commonalities of these programs.
This chapter concluded with a list of curriculum considerations for designing a
data science program.
 Fundamentals of Data Science for Future Data Scientists ◾ 193

References
Ackoff, R. L. (1989). From data to wisdom. Journal of Applies Systems Analysis, 16, 3–9.
Agarwal, R., & Dhar, V. (2014). Editorial—Big data, data science, and analytics: The oppor-
tunity and challenge for IS research. Information Systems Research, 25(3), 443–448.
Banafa, A. (2014). What is data science? Available at: http://works.bepress.com/
ahmed-banafa/15/.
Boston University Libraries. (n.d.). Data life cycle. Available at: https://www.bu.edu/
datamanagement/background/data-life-cycle/.
Buckland, M. K. (1991). Information as thing. Journal of the American Society for Information
Science, 42(5), 351–360.
Burtch Works Executive Recruiting. (2017). Data science salary study. Available at http://
www.burtchworks.com/big-data-analyst-salary/big-data-career-tips/.
Cao, I. (2017). Data science: Challenges and directions. The Communications of ACM, 60(8), 59–68.
Carter, D., & Sholler, D. (2016). Data science on the ground, hype, criticism, and every-
day work. Journal of the Association for Information Science & Technology, 67(10),
2309–2319.
Chang, C. (2012). Data life cycle. Available at: https://blogs.princeton.edu/
onpopdata/2012/03/12/data-life-cycle/
Cleveland, W. S. (2001). Data science: An action plan for expanding the technical areas of
the field of statistics. ISI Review, 69, 21–26.
Coronel, C., Morris, S., and Rob, P. (2012). Database systems: Design, implementation, and
management, 10th ed. Boston, MA: Course Technology, Cengage Learning.
DataScience.Community. (n.d.). Data science bootcamps. Available at: http://datascience.
community/bootcamps.
Davenport, H. T., & Patil, D. J. (2012). Data scientist: The sexiest job of the 21st century.
Harvard Business Review, 90(5), 70–77.
Davenport, T., & Prusak, L. (1998). Working knowledge: How organizations manage what
they know. Cambridge, MA: Harvard University Press.
Dhar, V. (2013). Data science and prediction. Communications of the ACM, 56(12), 64–73.
Doherty, R. (2010). Getting social with recruitment. Strategic HR Review, 9(6), 11–15.
Donoho, D. (2015). 50 years of Data Science. unpublished. Available at: https://dl.dropbox usercon-
tent.com/u/23421017/50YearsDataScience.pdf. University of Michigan’s Data Science Initiative.
Hawamdeh, S. (2003). Knowledge management: Cultivating knowledge professionals. Oxford,
UK: Chandos Publishing.
Hayashi, C. (1998). What is data science? Fundamental concepts and a heuristic example. In
C. Hayashi, K. Yajima, H. H. Bock, N. Ohsumi, Y. Tanaka, & Y. Baba (Eds.), Data
science, classification, and related methods. Springer, Tokyo: Studies in Classification,
Data Analysis, and Knowledge Organization.
Heimerl, F., Lohmann, S., Lange, S., & Ertl, T. (2014). Word cloud explorer: Text analytics
based on word clouds. Proceedings of the Annual Hawaii International Conference on
System Sciences (pp. 1833–1842). doi:10.1109/HICSS.2014.231.
Kandel, S., Paepcke, A., Hellerstein, J. M., & Heer, J. (2012). Enterprise data analysis and
visualization: An interview study. IEEE Transactions on Visualization and Computer
Graphics, 18(12), 2917–26. doi:10.1109/TVCG.2012.219.
194 ◾ Analytics and Knowledge Management

Laney, D. (2011). 3D data management controlling data volume, velocity, and variety.
META Group: Application Delivery Strategies, Available at: https://blogs.gartner.com/
doug-laney/files/2012/01/ad949-3D-Data-Management-Controlling-Data-Volume-
Velocity-and-Variety.pdf.
Leopold, G. (2017). Demand, salaries grow for data scientists. Datanami, January 24, 2017.
Available at: https://www.datanami.com/2017/01/24/demand-salaries-grow-data-
scientists/.
Losee, R. M. (1997). A discipline independent definition of information. Journal of the
American Society for Information Science, 48(3), 254–269.
Loukides, M. (2011). What is Data Science? Sebastopol, CA: O’Reilly Media.
Madden, A. D. (2000). A definition of information. Aslib Proceedings, 52(9), 343–349.
McKinsey & Company. (2011). Big data: The next frontier for innovation, competition, and
productivity. McKinsey Global Institute, (June), 156. doi:10.1080/01443610903114527.
Michigan Institute for Data Science (MIDAS). (2017). About MIDAS. Available at: http://
midas.umich.edu/about/.
Provost, F., & Fawcett, T. (2013). Data science and its relationship to big data and data-
driven decision making. Big Data 1(1), BD51–BD59.
Saracevic, T. (1999). Information science. Journal of the American Society for Information
Science, 50(12), 1051–1063.
SAS Institute. (2014). Getting Started with SAS ® Text Miner 13.2. Cary, NC: SAS.
Shum, S. B., Hall, W., Keynes, M., Baker, R. S. J., Behrens, J. T., Hawksey, M., &
Jeffery, N. (2013). Educational data scientists: A scarce breed. Available at: http://
simon.buckinghamshum.net/wp content/uploads/2013/03/LAK13Panel-Educ Data
Scientists.pdf.
Spirion. (n.d.). Data lifecycle management. Available at: https://www.spirion.com/
data-lifecycle-management/.
Stodder, D. (2015). Chasing the data science unicorn. Available at: https://tdwi.org/
articles/2015/01/06/chasing-the-data-science-unicorn.aspx.
Chapter 7

Social Media Analytics

Miyoung Chong and Hsia-Ching Chang

Contents
Introduction ......................................................................................................196
Historical Perspective of Social Networks and Social Media...............................199
Evolution of Analytics .......................................................................................201
Social Media Analytics.......................................................................................203
Defining Social Media Analytics ...................................................................203
Processes of Social Media Analytics .............................................................. 204
Social Media Analytics Techniques ....................................................................207
Identifying Data Sources...............................................................................207
Data Acquisition .......................................................................................... 208
Data Analysis Techniques............................................................................. 208
Sentiment Analysis .................................................................................. 208
Topic Modeling........................................................................................209
Visual Analytics ........................................................................................210
Stream Processing ..................................................................................... 211
Social Media Analytics Tools ............................................................................. 211
Scientific Programming Tools ....................................................................... 211
Network Visualization Tools .........................................................................212
Business Applications....................................................................................212
Social Media Monitoring Tools .....................................................................213
Text Analysis Tools........................................................................................213
Data Visualization Tools ...............................................................................213
Social Media Management Tools................................................................... 214
Representative Fields of Social Media Analytics ................................................. 214
Conclusions ...................................................................................................... 215
References .........................................................................................................216

195
196 ◾ Analytics and Knowledge Management

Introduction
Social media analytics (SMA) has become one of the core areas within the exten-
sive field of analytics (Kurniawati, Shanks, & Bekmamedova, 2013). Generally
speaking, SMA facilitates proper analytic techniques and tools to analyze user-
generated content (UGC) for a particular purpose. The content of social media
comes in different forms of repositories (Sinha, Subramanian, Bhattacharya, &
Chaudhary, 2012). For instance, the repositories could be blogs (Tumblr), microb-
logs (Twitter), wikis (Wikipedia), social networking sites (Facebook and LinkedIn),
review sites (Yelp), and multimedia sharing sites (YouTube) (Holsapple, Hsiao, &
Pakath, 2014).
SMA has involved many techniques and tools to mine a variety of UGC,
including sentiment analysis, topic modeling, data mining, trend analysis, and
social network analysis. A primary reason for growing interest in SMA is the real
time accessibility regarding size and diffusion speed of the UGC on social media
(Holsapple et al., 2014). For example, a YouTube video showing a Coke bottle
exploding because of Mentos became the most popular video on YouTube and
subsequently in news shows in 2009 (Kaplan & Haenlein, 2010). Despite the bur-
geoning attention to SMA, social media analysts need to deal with arduous tasks
in the given context to accomplish a specific analytical objective because the UGC
from social media is generally improvised, freeform, and a mixture of relevant and
irrelevant resources.
During the 1990s, the Internet and the World Wide Web (WWW) were
adopted to facilitate social communication. The development and quick diffusion
of Web 2.0 technologies made a revolutionary leap in the social element of the
Internet application. Social media users can take advantage of the user-centered
platforms with UGC while the various set of possibilities exist to connect these
cyberspaces to construct online social networks. Adopting social media platforms
serves diverse purposes, such as social interaction, marketing, digital education,
disaster management, and civic movement, and by user groups, including business,
governments, nongovernment organizations, politicians, and celebrities. For exam-
ple, Psy’s music video, Gangnam Style released on July 15, 2012, became the most
watched video on YouTube history with the view of more than 2.8 billion times,
which also tremendously contributed to making the singer to become a world star.
Facebook, starting its service in 2004, currently reached more than 2 billion
active users worldwide as of June 2017. Launched in 2006, Twitter has obtained
319 million active users as of April 2017 and created 500 million Tweets every day.
Though Twitter has a smaller number of active accounts than YouTube, WhatsApp,
or Facebook Messenger, Facebook and Twitter have been the most visible social
media platforms initiating new functional services by incorporating Web 2.0 com-
ponents into other web-based applications (Obar & Wildman, 2015). When com-
pared to the lists in 2015 and 2017, overall the popular social media platforms are
similar besides the leap of WhatsApp and Facebook Messenger. Figure 7.1 presents
 Social Media Analytics ◾ 197

1.4
1.3
1.2
1.1
1.0
0.9
Percent growth

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
−0.1
m

ha bo

Sn pp

Q a
ne
Tw ok

e
r

W at

Fa hat

er

b
bl

Ba kyp
ra

Q
h

itt

tie
zo
ei
A

bo
m

eC

c
ag

ts

ap

S
Tu

ce

u
st

id
na
In

W
Si

Figure 7.1 The percentage growth between 2015 and 2017 among social media
services that have most active users.

the significant growth of users of Tumblr, Instagram, WeChat, Sina Weibo, and
WhatsApp between 2015 and 2017.
An official definition of social media requires distinguishing the two interre-
lated concepts, Web 2.0 and UGC (Kaplan & Haenlein, 2010). The term Web 2.0
was first adopted in 2004 to describe the new web environment where software
developers and end users began to mash up the content and applications coopera-
tively and innovatively. Though Web 2.0 does not indicate any particular technical
term related to the WWW, it includes a few essential functionalities for its perfor-
mances such as Adobe Flash, Really Simple Syndication (RSS), and Asynchronous
JavaScript (AJAX) (Kaplan & Haenlein, 2010).
While considering Web 2.0 as the platform of the social media development and
technological foundation, UGC refers to the collective outcomes of user interac-
tions with social media. Extensively employed in 2005, UGC represents the diverse
formats of content in social media (Kaplan & Haenlein, 2010). The Organization
for Economic Cooperation and Development specified three requirements of UGC:
public accessibility of the content, creativity, and amateurism (Peña-López, 2007).
Drawing upon these descriptions about Web 2.0 and UCG, Kaplan and Haenlein
(2010) define social media as a “group of Internet-based applications that build on
198 ◾ Analytics and Knowledge Management

the ideological and technical foundations of the Web 2.0, and that allow the cre-
ation and exchange of User Generated Content” (p. 61). Extending this perspective,
Batrinca and Treleaven (2015) delineate social media as “web-based and mobile-
based Internet applications that allow the creation, access and exchange of UGC
that is ubiquitously accessible” (p. 89).
To characterize social media, Kaplan and Haenlein (2010) categorize social
media platforms by two dimensions: social presence and media richness, and self
presentation and self disclosure. Table 7.1 presents the six different types of social
media. Regarding social presence and media richness, collaborative social media
applications, such as Wikipedia, blogs represent the lowest scores because these are
usually text-based services and limited to simple trade regarding content sharing.
The middle level presents content communities, such as YouTube, and social
networking sites, such as Facebook and Instagram, which allow sharing images
and multimedia content as well as text-based communication. The highest level
includes virtual game and social worlds, such as World of Warcraft, Second Life,
and virtual reality role-playing games, which pursue any possible aspects of direct
interactions in a cyber world (Kaplan & Haenlein, 2010). As for self-presentation
and self-disclosure, blogs ranked higher than collaborative projects because they are
more focused on particular content areas. Likewise, social networking sites present
more self-disclosure than content communities. Lastly, cyber social worlds demand
further self-disclosure than virtual game worlds because disclosers are strictly
bounded by game rules (Kaplan & Haenlein, 2010).
Obar and Wildman (2015) focus on the importance of user profile elements in
defining social media. Despite the substantial differences in the options of identi-
fying users and information requested depending on the social media platforms,
they usually require creating a user name along with contact details and a profile
picture. The requirement enables social media users to make connections and share
their content without many doubts because without verifying user information,
discovering and linking to other users could be a challenge. Obar and Wildman
(2015) describe the functions of social media platforms as connecting the online

Table 7.1 Classification of Social Media by Social Presence/Media Richness

and Self-Presentation/Self-Disclosure
Social Presence/Media Richness

Self-presentation/ Low Medium High

Self-disclosure
High Blogs, Quora, Facebook, Second life
Flicker Instagram

Low Wikis YouTube, World of Warcraft,

Pinterest virtual reality
role-playing game
Source: Kaplan, A. M., and Haenlein, M., Bus. Horiz., 53, 59–68, 2010.
 Social Media Analytics ◾ 199

social networks. Along this line, boyd and Ellison (2010) considered user profiles as
the backbone of social media platforms.
A majority of Americans now state that they obtain news more increasingly via
social media (Kohut et al., 2008). According to a recent study, 60% of all Americans
use social media to get news (Gottfried & Shearer, 2016), whereas approximately
80% of online Americans are currently Facebook users and among these users, 66%
obtain news on the site (Greenwood, Perrin, & Duggan, 2016). On Twitter, around
60% of the users receive news on the site, which results in a bigger percentage with
a smaller user foundation (16% of American adults). In addition, research studies
have continuously discovered that the more people read news media, the more likely
they are civically and politically involved across various measures (Wihbey, 2015).

Historical Perspective of Social Networks

and Social Media
Throughout history, humans have developed technologies to communicate with
each other. There are plenty of ideas regarding the first advent of social media, but
the adoption of the telegraph to exchange messages in 1792 can be one of the earli-
est examples social media (Edosomwan, Prakasan, Kouame, Watson, & Seymour,
2011). In the late 1800s, drawing from the social contract conceptions of society,
German sociologist Ferdinand Tonnies explained that social groups have formed by
sharing values or conflicts, and Durkeim, a French sociologist, applied sociological
theory to empirical research. People used the radio and telephone for social interac-
tion, although the radio communication was one-sided (Rimskii, 2011).
Social networks have evolved over the decades by employing digital media in the
modern day. However, social communication using technology started with the tele-
phone system in the 1950s. Phone phreaks used handcrafted devices to access the tele-
phone network for both exploration and to exploit vulnerabilities allowing free access
(Borders, 2009). E-mail was introduced to the public in the 1960s. In 1969, ARPANET,
a network of time-sharing computers that established the foundation of the Internet cre-
ated by a U.S. government agency, Advanced Research Projects Agency, was developed
(Edosomwan et al., 2011). In the same year, CompuServe, an early form of cloud com-
puting, was designed to provide time-based services by leasing time on its machine, but
this was ineffective because of the high fees for the service (Rimskii, 2011).
During the 1970s, the idea of social media was developed through computer tech-
nologies. For example, Multi-User Dungeon (MUD), a real time and online role-play-
ing game organized through interactive fiction, is mainly text-based, where users need
to input commands in a natural language. Bulletin board system (BBS) was introduced
in 1978, and it allows its users to upload and download software, trade messages, and
read news by logging into the system. Bulletin boards were the precursor of the WWW.
The Usenet, developed by professors at Duke University in 1980, is an online discus-
sion system where users can post news, messages, and articles (Ritholz, 2010).
200 ◾ Analytics and Knowledge Management

During the 1980s, the Whole Earth Lectronic Link (WELL), General Electric
Network for Information Exchange (GEnie), Listserv, and IRC were introduced.
The WELL started in 1985 as a conversation-based community through BBS and
is one of the long-standing online communities. GEnie is an Internet application
programmed with ASCII language and was in a competitive relationship with
CompuServe. Launched in 1986, Listserv was the first electronic mailing applica-
tion through software installed on the server, and it enabled automatic and machine-
processed electronic emailing distribution, which allowed an e-mail to reach a group
of people registered on the mailing list. IRC (Internet relay chat) was created for
group conversations as a type of real time chat, online text messaging, concurrent
conferencing, and data transferring between two people (Ritholz, 2010).
During the 1990s, many social networking sites, such as SixDegrees.com,
MoveOn.org, BlackPlanet.com, and AsianAve.com, were created. These web-
sites were Internet niche social networking sites with which users can interact.
Additionally, blogging sites were created, such as Blogger and Epinions where
users can find and write review comments about commodities (Edosomwan et al.,
2011). In this period, ThirdVoice, a free plug-in user-commenting service on web
pages, and Napster, a peer-to-peer music file sharing software application, were
generated. Opponents of ThirdVoice criticized that the comments were frequently
insulting. Through Napster, users could exchange music files by deviating from
the legal distribution methods; this eventually determined the end user to be a
violator of copyright laws (Ritholz, 2010).
In 2000, many social networking sites sprang up, including LunarStorm, Cyworld,
and Wikipedia. LunarStorm (www. LunarStorm.se), a Swedish commercial virtual site
and Europe’s first Internet community, was specially designed for teenagers. In 2001,
Fotolog, sky blog, and Friendster started their services. In 2003, MySpace, LinkedIn,
tribe.net, Last.fm and in 2004, Facebook Harvard, Mixi, and Dogster emerged.
In 2005, Yahoo! 360, BlackPlanet, and YouTube evolved (Junco, Heibergert, &
Loken, 2011). Though BlackPlanet was created in the 1990s, it became popular and
evolved in 2005. Twitter launched its service in 2006, and the number of users rapidly
increased because of its microblogging style and celebrity adoption (Jasra, 2010).
2010 was Facebook’s year. Facebook took over Google’s position as in the big-
gest website regarding market share in July 2010, and Time magazine recognized
Facebook CEO Mark Zuckerberg as Man of the Year (Cohen, 2010). In the Harvard
Business Review, Armano (2014) describes six social media trends for 2010. He
points out that there will be more scaled social initiatives beyond one-off market-
ing or customer relations efforts. For instance, Best Buy’s employees were directed
to participate in customer support on Twitter through a company-built system that
monitors their participation. With approximately two-thirds of organizations pro-
hibiting access to social network applications from corporate-owned devices, while
simultaneously the sales of smartphones have skyrocketed, employees are likely to
feed their social media cravings on their mobile devices (Armano, 2014). Sharing
content through online social networks rather than by email became mainstream
 Social Media Analytics ◾ 201

for people. For example, the New York Times created a content sharing application
that enables them to quickly distribute an article across social media networks such
as Facebook and Twitter (Armano, 2014).
Communication methods, such as face-to-face meetings, phone calls, and
email, have restrictions. For example, people can easily forget and lose the content
of a conversation because of memory loss or missing notes. Social media aids inter-
action and communication among individuals and helps them to reach a larger
audience. Adopting social media has effectively raised the channels of communica-
tion (Edosomwan et al., 2011) because it has become much easier for people to send
timely messages via a Tweet or an instant messenge, which saves time and energy
(interacting with people using social media technology). In addition, engaging
social media reinforces brand experience, which helps to establish a brand reputa-
tion. Customer brand awareness begins with the employees’ experiences about the
company (Carraher, Parnell, & Spillan, 2009).

Evolution of Analytics
The area of analytics started in the mid-1950s along with new methods and tools
that created and captured a significant amount of information and distinguished
patterns in it faster than unaided human intelligence ever could. The concepts and
techniques of business analytics have continuously changed based on analytical
circumstances. The era of Analytics 1.0 can be described as the era of “business
intelligence (BI).” In this period, an analytic technique made progress on providing
an impartial comprehension of major business developments and provided manag-
ers with fact-based understanding for better insights to make marketing decisions.
From this period, production data, including processes, sales, and customer care,
was recorded, accumulated, and innovative computing techniques played a critical
role. For the first time, customized information systems with considerable scale
increased the investment to the systems. Analysis process required far more time to
have data than to analyze it. The process, which took weeks to months to conduct
analysis, was slow and meticulous. Therefore, it was critical to discover the most
important questions and concentrate on them. In this era, analytics focused on
better decision-making opportunities to enhance predictions on specific primary
components.
The first generation of analytics was followed by the era of Analytics 2.0, known
as the era of Big Data. Analytics 1.0 continued until the mid-2000s, when inter-
net-based companies, such as Google, Facebook, Twitter, and LinkedIn, started
to collect and analyze new types of information (Davenport & Court, 2014). In
this Analytics 2.0 phase, the industry adopted new tools to have innovative ana-
lytic techniques and attract customers. For instance, LinkedIn built many big data
products, such as Network Updates, People You May Know, Groups You May Like,
Jobs You May Be Interested In, Companies You May Want to Follow, and Skills
202 ◾ Analytics and Knowledge Management

and Expertise (Davenport & Court, 2014). This improvement required powerful
infrastructure and analytics talents who can develop, obtain, and master different
types of analytics tools and technologies. A single server could not perform analysis
of big data.
Hadoop, an open-source parallel server software structure for fast batch data
processing, was introduced. A NoSQL database system was employed to manage
relatively unstructured data. The cloud-computing system was adopted to man-
age and analyze lots of information. Machine-learning (ML) techniques were also
applied to quickly create models from the rapidly-shifting data. Data analysts
had to make it organized for statistical analysis with scripting languages, such as
Python, Hive, and Pig. Spark, an open source cluster computing system especially
for streaming data, and R, a framework for statistical computing and graphics,
have become popular as well. The required capabilities for Analytics 2.0 were much
higher than those for Analytics 1.0. The quantitative analysts of Analytics 2.0 were
named data scientists, who had both computational and analytical skills.
The revolutionary companies in Silicon Valley started to pay attention to ana-
lytics to promote shopper-facing products and services. They drew consumers’
attention to their websites through certain computational algorithms, recommen-
dations from other users, product reviews, and targeted advertisements, which are
driven by analytics drawn from enormous quantities of data.
When analytics entered the 3.0 phase, almost every company in every industry
makes decisions on products and services based on data analytics. An enormous
amount of data has been created whenever companies produce, deliver, and do
marketing for products and services or interact with customers. For example, UPS
entirely reorganized its delivery routes by implementing big data analytics. UPS
has traced and recorded package shipment and delivery since the 1980s by accu-
mulating information on the 16.3 million (on average) package transactions and
movements daily, which obtains 39.5 million daily tracking requests. The telemat-
ics sensors installed in more than 46,000 UPS trucks trace metrics such as speed,
braking, direction, and training performance. This tracked data demonstrates daily
performance as well as information about a substantial reconstruction of delivery
routes. UPS’s On-Road Integrated Optimization and Navigation (ORION) is a
proprietary route optimization software that relies greatly on online map data and
optimization algorithms.
This analytics solution will eventually be able to redesign a UPS driver’s pick-
ups and distributions in real time. It is also known as “the world’s largest opera-
tions research project” due to its scale and scope (Davenport & Court, 2014). In
2015, UPS adopted ORION to 70% of U.S. routes identified as part of the initial
deployment, which aided UPS to diminish driving miles, fuel consumption, and
car emissions. With the full employment of the ORION system in 2016, UPS
expected to save $300–$400 million by saving 10 million gallons of fuel, reducing
100 million miles, and reducing 100,000 metric tons of carbon dioxide production
annually, which is equal to taking more than 20,000 passenger vehicles off the road
 Social Media Analytics ◾ 203

each year. ORION costs $250 at full deployment, but ORION has already saved
UPS more than $320 million at the end of 2015 (Informs.org, 2017).
The ORION project demonstrates that a company’s application of analytics
not only improved internal business decisions but also generated values, including
environmental protection. This is the core of Analytics 3.0. Companies employed
analytics to assimilate information and provide examples while hinting at other
possible applications. Internet-based firms, such as Google, Facebook, Amazon,
eBay, and Facebook, processing enormous amounts of streaming data, have become
frontiers of this approach. These online companies have bloomed mainly by assist-
ing their customers to make effective decisions and actions, and not by providing
simple information for them. This strategic shift signifies an advanced function for
analytics within institutions. Companies need to identify a division of related tasks
and react to new competence, environments, and preferences.

Social Media Analytics

Social media has dramatically changed information creation and sharing. What
was once the exclusive domain of traditional news media, television, radio, public
institutions, and private entities are willing to reach out to social media. Thanks
to the prevalence of social media, information dissemination on a mass scale now
flows from a customer’s fingertips to the Internet. According to Chung (2016),
“Its key capabilities include universal reachability, speed of interaction, multimodal
communication, device compatibility, and emotional appeals of SM” (p. 105).
Nearly ubiquitous Internet access combined with intuitive web applications has
democratized mass media. Social media supports many forms of Internet-fueled
communication. Therefore, an individual can rate any customer experience (review
websites), maintain an electronic diary (weblog or blog), provide recommendations
for product purchase (recommender website), and communicate privately with
friends and family or publicly with strangers around the world through different
social media platforms.

Defining Social Media Analytics

Regardless of medium, social media outlets capture tremendous amounts of data.
The data are a treasure trove awaiting discovery. In recognition of the underlying
value embedded in social media data, analysts have begun to apply the power
of data analytics to social media data. While sharing analytic techniques with
other forms of media, SMA focuses exclusively on various types of data from
social media. SMA attempts to unveil patterns of unstructured data by develop-
ing and evaluating tools that can capture, analyze, synthesize, and visualize social
media data (Zeng, Chen, & Li, 2010). From the perspective of business, Holsapple
et al. (2014) define business SMA as “all activities related to gathering relevant
204 ◾ Analytics and Knowledge Management

social media data, analyzing the gathered data, and disseminating findings as
appropriate to support business activities such as intelligence gathering, insight
generation, sense-making, problem recognition/opportunity detection, problem
solution/opportunity exploitation, and/or decision-making undertaken in response
to sensed business needs” (p. 4).

Processes of Social Media Analytics

Regarding the processes of performing SMA, Fan and Gordon (2014) describe
SMA as a three-step process progressing mostly linearly (some steps overlap) from
capture to understand and ending with presentation. During capture, data are
gathered and preprocessed, and pertinent information is extracted. In the under-
standing phase, data are further analyzed using advanced analytics techniques,
such as sentiment analysis, topic modeling, social network analysis, trend analysis,
and so on. During the presentation phase, the analysis is summarized and visual-
ized as a series of reports. Bengston et al. (2009) view SMA as a monitoring process
consisting of five phases: the definition of the problem, the identification of online
news sources that will be used to collect content, the development of search terms
with respect to several strategies and the storage of search results, the analysis of
textual content as the core phase of the analytics process, and the presentation of
gained results.
Another approach is proposed by Bruns and Liang (2012) for the analysis of
Twitter data. They suggest a three-step approach including data collection and stor-
age, data analysis, and the publication of results. Similarly, Stavrakantonakis et al.
(2012) mention four steps starting with the collection of data, the establishment of
a “listening grid” to store gathered data, followed by the actual analysis, and finally
the generation of “actionable reports” to support decision-making (p. 54). Likewise,
Ruggiero and Vos (2014) present a four-step approach, which are preparation, data
collection, data analysis, and reporting. Summarizing the existing procedures of
an analytics pipeline, Wittwer, Reinhold, and Alt (2016) suggest four main steps.
First, the definition of media sources and appropriate search terms. Second, the
pre-processing of data for integrating and aligning the heterogeneous data from
different sources as well as the application of various analysis techniques. Third,
the presentation and interaction of users with the results. Fourth, the use of results,
which is the only manual task (Wittwer et al., 2016).
Social media data are roughly categorized as historical datasets and real-time
feeds. Data is further classified as raw, cleaned, and value-added. Historical data
are archives of previously captured news feeds, financial and economic data, and
social commentary, whereas real-time feeds are continuous data streams taken
directly from a social media site, news feed, or financial exchanges (Batrinca &
Treleaven, 2015). Social media data is acquired (captured) through a variety of
techniques. Some data providers (e.g., Wikipedia) allow a full database down-
load. Other vendors provide controlled access to data via dedicated software
 Social Media Analytics ◾ 205

tools (e.g., Google Trends). Lastly, data are available through an application
programming interface (API). An API exposes data over HTTP for system-
to-system data transfer. Examples include Wikipedia’s DBpedia, Twitter, and
Facebook (Batrinca & Treleaven, 2015).
Raw data are the data as the source creates it. It may contain errors and may
have never been fixed. Cleaned data are raw data that has had some amount of pre-
processing to remove errors such as typos, wrong facts, outliers, or missing informa-
tion. Value-added data are cleaned data that have been augmented with additional
knowledge gleaned from the analysis (Batrinca & Treleaven, 2015).
As for data formats, data is most commonly encoded into hypertext markup
language (HTML), extensible markup language (XML), JavaScript object nota-
tion (JSON), or comma-separated value (CSV) files. HTML is a markup language
for authoring web pages. It is responsible for delivering page structure directives
(tables, paragraphs, and sections) and content (text, multimedia, and images) to
a web browser. XML is a markup language used to create structured textual data.
XML markup functions as metadata that describes the content contained within
the markup. Both XML and HTML wrap content between start and end tags
(e.g., </div>). JSON is a lightweight data structure derived from a subset of the
JavaScript programming language. It has a simple key-value structure that is simul-
taneously a machine and human readable format. Because JSON is familiar to
C language software developers, it is used in many data feeds. CSV refers to any
file that has a single record per line, fields that are separated by a comma, and text
that is encoded using ASCII, Unicode, or EBCDIC (Batrinca & Treleaven, 2015).
After data are captured, analysts can proceed to the understanding phase. In
this step, the actual value of SMA is revealed. Value is defined through the lens
of the analyst. An analyst working under the auspices of a corporation has a dif-
ferent concept of value than an analyst working to support a political candidate.
Holsapple et al. (2014) offer the following benefits of business SMA: improved
marketing strategy, better customer engagement, better customer service, reputa-
tion management, and new business opportunities. Like epinion.com and Amazon,
review websites allow customers to post comments about an experience with a
product or service. SMA can provide insights regarding these experiences that in
turn inform marketing strategy.
Customer engagement improves when businesses target customer values and
provide additional channels for two-way communications. Customer service
improves as companies become more in tune with the needs of their clients. As
an example, SMA identified stock picking experts on the Motley Fools CAPS
voting site. Picks by these experts consistently outperformed other voters. These
stock picks were provided to Motley Fool customers creating a better product
and improving customer service. SMA assists with reputation management
by allowing a business to extract sentiment insight from social media content
related to the company. Finally, SMA can identify new business opportunities by
monitoring distinctive phrases that spread rapidly on social media applications.
206 ◾ Analytics and Knowledge Management

These phrases offer insights that can help a firm’s decision on providing a new
product or what new features to add to an existing product. That is because
SMA has played a role in each stage of the product development lifecycle: design
development, production, utilization, and disposal (Fan & Gordon, 2014). In the
design development phase, trend analysis can identify a change in customer atti-
tude and desires for a product. Application development firms regularly release
multiple product versions and then solicit feedback directly from potential end
users. Responses then drive the next product feature sets. In extreme cases, end
users and software developers cocreate products using social media as a collabora-
tion platform. In the production phase, SMA allows firms to anticipate changing
demand by scaling production either up or down. A company can also use social
media data to track issues posted by competing companies regarding a supplier,
thus allowing the company to avoid the same problem.
Furthermore, SMA is most commonly leveraged during the product adoption
phase. Firms closely monitor brand awareness (introduction to the brand), brand
engagement (connection to the brand), and word of mouth (customer chatter about
a brand). Metrics, like the number of Tweets and followers, indicate an excite-
ment (positive or negative) for a brand. Customer segmentation (grouping custom-
ers based on shared attributes) allows a firm to tailor marketing messages for a
particular group. In the product disposal stage, SMA can follow, and companies
can directly communicate with, consumers on disposal strategy. This is especially
useful when product disposal creates an environment risk (e.g., toxic batteries).
Because product disposal is often accompanied with product replacement, firms
can use SMA to market replacement products selectively.
Similar to people in business, politicians benefit from SMA as well. The ben-
efit, however, is different as politicians have wildly divergent needs. Reputation
and impression management are of utmost importance politically. Politicians are
interested in how people discuss them, new topics that might trigger crises or
sway sentiment against them, and measuring the degree of influence they exert.
Additionally, SMA is applied in an exploratory fashion to monitor for “late-
breaking” topics thus gaining advanced notification and a longer preparation
period (Stieglitz & Dang-Xuan, 2013). Politicians use social media to gain sup-
port, encourage civic engagement, and promote themselves. President Obama’s
2008 campaign is widely considered the first campaign where social media and
SMA had a measurable and significant effect on the election outcome (Grubmüller,
Götsch, & Krieger, 2013).
Stieglitz and Dang-Xuan (2012) propose an analytics framework for con-
ducting SMA in the political realm consisting of two major parts: data tracking
and monitoring, and data analysis. Data tracking and monitoring are concerned
with techniques for acquiring social media data while data analysis pertains to
analysis methods and reporting. For data analysis, they suggest three approaches:
topic or issue-related, opinion or sentiment-related, and structural. Topic or issue-
related analysis or issue management uses text mining techniques to identify topics
 Social Media Analytics ◾ 207

(issues) that might become a crisis or scandal and damage reputation. Opinion or
sentiment-related analysis is an attempt to identify citizen (voter) opinion regarding
a topic. Politicians use such information to make informed decisions about upcom-
ing votes, which issues to address or refute, and decision-making. The structural
analysis attempts to identify key players in a community network. This information
allows a politician to seek favor with influential people. The same approach can get
extended to entire communities if it appears that individual communities exert
more influence on an issue.

Social Media Analytics Techniques

This section introduces SMA techniques pertaining to different procedures of SMA.
Here we reiterate the common processes of SMA specified in relevant studies: iden-
tifying data sources, data acquisition, data analysis, and representing results.

Identifying Data Sources

Data sources could be solely textual content or aggregating multiple data sources,
including social networking services, blogs, and RSS feeds, supplemented by vari-
ous types of data, such as geospatial data, video data, and communication data
(Batrinca & Treleaven, 2015). Depending on looking backward or forward, we can
group social media data into:

◾ Historical data: Recently amassed and stocked social, news, and business-
related data.
◾ Real-time data: Streaming data from social media, news agencies, and mone-
tary trade, telecommunication services, and global positioning system (GPS)
gadgets.

Data also subdivides into raw, cleaned, and value-added data based on the level of
the process. Raw data are primary and unprocessed data directly from the source
that includes errors. Cleaned data is processed and modified raw data by removing
erroneous parts. Value-added data are cleaned, analyzed, tagged data, or augmented
data with information. Analysts need access to historical and real-time social media
data, particularly the primary sources, to conduct comprehensive research. In many
cases, it is beneficial to combine different data sources with social media data. For
example, opinions about negative financial news, such as a downslide of the stock
market, might be presented in social media. Thus, when conducting textual data
analysis of social media, considering multiple data sources will potentially help
detect the context and perform deeper analysis with a better understanding of
the data. The aggregation of data sources is certainly the focus of future analytics
(Batrinca & Treleaven, 2015).
208 ◾ Analytics and Knowledge Management

Data Acquisition
There are four ways to directly acquire data from social media, including scraping,
application programmable interfaces (API), RSS feed, and file-based data acquisi-
tion. Scraping is gathering social media data from the Internet, and this is usually
unstructured text data. Scarping is also known as site scraping or web data extrac-
tion, web harvesting, and web data mining. Analysts can collect social media data
systematically if social media data repositories guarantee programmable HTTP-
based approach to the data through APIs. Facebook, Twitter, and Wikipedia pro-
vide access via APIs. An RSS feed is a systematic method of streaming social media
data to deliver to its subscribers. An RSS feed is considered semi-structured data.
File-based data is usually acquired from spreadsheets and text files.

Data Analysis Techniques

Sentiment analysis, topic modeling, and visual analytics are the primary analysis
techniques of social media data. Sentiment analysis, namely opinion mining, is usu-
ally employed to determine the attitude and perception of an author concerning an
event. For this purpose, natural language processing (NLP), computational linguis-
tics, and text analytics are deployed to identify and extract subjective information
from textual data. Two quantitative methods of topic modeling, latent Dirichlet
allocation (LDA) and latent semantic analysis (LSA), facilitate unsupervised (unla-
beled) text analysis and can also be implemented to analyze social media data.

Sentiment Analysis
Web 2.0 users can form, participate in, and collaborate with virtual online commu-
nities. Microblogs, such as Twitter, became user-generated information-abundant
resources, because users began sharing information and opinions in diverse online
events and domains, ranging from breaking news to celebrity gossip, to product
reviews, and to discussions about recent incidents, such as the Orlando massacre in
June 2016, the U.S. Presidential election, and hurricanes Harvey and Irma in 2017.
Social media includes a copious amount of sentiment-embodied sentences.
Sentiment refers to “a personal belief or judgment that is not founded on proof
or certainty” (WordNet 2.1 definitions), which may depict the emotional state of
the user, such as happy, sad, angry, or the author’s viewpoint on a topic. Sentiment
analysis, an important aspect of opinion mining, aims to discover whether the
polarity of a textual corpus, a collection of written texts, leans towards positive,
negative, or neutral sentiments.
During the past decade, sentiment analysis has been a popular research area in
SMA, particularly on Twitter because of the accessibility to diverse fields. Scholars
have applied sentiment analysis to business predictions (Liu, Huang, An, & Yu,
2007), politics (Park, Ko, Kim, Liu, & Song, 2011), finances (Dergiades, 2012),
 Social Media Analytics ◾ 209

and the automatic identification of sentiment on textual corpora (Kontopoulos,

Berberidis, Dergiades, & Bassiliades, 2013). Sentiment analysis can be conducted
at different levels, such as the document, sentence, word, or feature level. In the
case of Twitter analytics, word-level analysis fits better because Twitter limits users
to 140 characters of information. Cross-examination of the literature verifies that
the evaluation measures of mechanically annotating sentiment at the word level lie
in two approaches: “dictionary-based approaches” and “corpus-based approaches”
(Kumar & Sebastian, 2012, p. 372). Moreover, for automated sentiment analysis,
diverse methods have been adopted to forecast the sentiment of words, articulations,
or corpora, and these approaches incorporate NLP and ML algorithms (Kumar &
Sebastian, 2012). Recently, researchers analyzed Twitter activities and reported that
more than 80 percent of users either post status updates to their followers or spread
information regarding their daily experiences (Thelwall, Buckley, & Paltoglou, 2011).
Sentiment analysis is about mining attitudes, emotions, feelings—it is subjec-
tive impressions rather than facts. Generally speaking, sentiment analysis aims to
determine the attitude expressed through the text with respect to the topic or the
overall contextual polarity of a document (Mejova, 2009). Pang and Lee (2008) pro-
vide a thorough documentation on the fundamentals of sentiment classification and
extraction, including sentiment polarity, degrees of positivity, subjectivity detection,
opinion identification, nonfactual information, term presence versus frequency,
parts of speech (POS), syntax, negation, topic-oriented features, and term-based fea-
tures beyond term unigrams. Moreover, sentiment analysis divides into sentiment
context, sentiment level, sentiment subjectivity, sentiment orientation (polarity),
and sentiment strength depending on specific tasks. Used for discovering the con-
text of the text to identify opinion, sentiment context varies from expert review
portals to generic forums where opinions cover a spectrum of topics (Westerski,
2008). Sentiment level means that the text granularity of sentiment analysis can be
performed, such as the document, sentence, or feature level. Sentiment subjectiv-
ity determines whether a given text conveys an opinion or states a fact. Sentiment
orientation (polarity) identifies whether a viewpoint in a text is positive, negative or
neutral. Sentiment strength discovers the “degree” of an idea in a given text.

Topic Modeling
According to Blei, Ng, and Jordan (2003), “topics” are probability distributions
across all terms in the dictionary. For example, the topic “education” can be associ-
ated with terms such as “students,” “teachers,” “schools,” and so on, with high prob-
ability, and the likelihood of the rest of the words in the dictionary—which typically
will include hundreds or thousands of terms—will be near 0. LDA assumes that the
author of a document produces text by following a generative model, according
to which, given the document, first a topic is selected from a corresponding con-
ditional multinomial distribution of topics, and then, given the topic, words are
chosen from the multinomial distribution of terms that corresponds to that topic.
210 ◾ Analytics and Knowledge Management

Since the problem of estimating the parameters of the multinomial (Dirichlet) dis-
tributions of documents across topics and topics across terms is intractable, they are
estimated using Markov chain Monte Carlo simulations. LDA-based topic modeling
has two advantages. First, utilizing clues from the context, the topic models connect
words with related meanings and separate uses of words from multiple meanings
(McCallum, 2002). Second, since the topic modeling method is performed auto-
matically based on a mathematical algorithm, subjective bias in analyzing data is
minimized. A tutorial for LDA is available in ConText (Diesner et al., 2015).
LSA facilitated topic extraction as well as document similarity and was introduced
as a novel way of automatic indexing and information retrieval in library systems
in the early 1990s (Deerwester, Dumais, Furnas, Landauer, & Harshman, 1990).
According to Deerwester et al. (1990), “this approach takes advantage of implicit
higher-order structure in the association of terms with documents (‘semantic struc-
ture’) in order to improve the detection of relevant documents on the basis of terms
found in queries” (p. 1). LSA has been extended as a theory of meaning where topics
are represented in two equivalent forms: as linear combinations of related terms and
as linear combinations of relevant documents (Foltz, Kintsch, & Landauer, 1998). As
a statistical estimation method, LSA quantifies a collection of records by applying the
vector space model (VSM), which arranges text-based data into a term-by-document
matrix where term frequencies are recorded. LSA then employs singular value decom-
position (SVD), which is an extension of principal component analysis and quantifies
patterns of term-document co-occurrence using least squares estimation. After SVD,
an analysis similar to what is done in numerical factor analysis can produce interpre-
table topics (Evangelopoulos, Zhang, & Prybutok, 2012; Sidorova, Evangelopoulos,
Valacich, & Ramakrishnan, 2008). Despite the more rigorous statistical estimation
in LDA, LSA has higher computational efficiency, provides reproducible results, and
is readily available in several implementation packages (Anaya, 2011).

Visual Analytics
Visual analytics (Cook & Thomas, 2005) draws on methods from information
visualization (Card et al., 1999) and computational modeling. Visual analytics has
considerably contributed to SMA (Diakopoulos et al., 2010; Hassan et al., 2014).
The theory of visual analytics employs interactive visualization in processing ana-
lytical reasoning instead of the static display as the results of the analysis (Cook &
Thomas, 2005). The analytic process starts with a high-level assessment that leads
analysts to interesting features of the data. Then the analysts can reconstruct the
perspective by filtering or generating brand-new visualizations, which help them
explore better via the qualitative data analysis.
Brooker, Barnett, and Cribbin (2016) address the advantage of visual analytic
approach to social media data. By integrating data collection and data analysis
as a single process, exploring a dataset may inspire innovative ideas that eventu-
ally result in new levels of data collection with findings discovered during the
 Social Media Analytics ◾ 211

continuing iterations (Brooker et al., 2016). Therefore, through visual analytics,

data analysts can independently examine their hypotheses and viewpoints to dis-
cover phenomena as they develop questions to assess their ability in mapping digital
data to traditional social science concepts (Brooker et al., 2016).

Stream Processing
Data analytics of real-time social media characterizes large quantity of temporal data
with little latency. This process requires applications that support online analysis of
data streams. However, traditional database management systems (DBMSs) lack the
pre-established concept of time and cannot manage online data in real time, which
results in developing data stream management systems (DSMSs) (Hebrail, 2008).
DSMSs can process main memory while saving the data on the system, which enables
it to deal with the transient online data streams and to process constant queries of
the streaming data (Botan et al., 2010). Commercial DSMSs includes CEP engine
(Oracle), StreamBase, and StreamInsight (Microsoft) (Chandramouli et al., 2010).
Taking Twitter as an example, Tweets generated from public accounts that represent
more than 90% of Twitter accounts including replies and mentions, can be retrieved
in JSON format. For example, the Twitter Search API is used to request past data
on Twitter and streaming API, filtered by user ID, search keyword, and geospatial
location, is used to request a real time stream of Tweets (Batrinca & Treleaven, 2015).

Social Media Analytics Tools

Many opinion mining tools have been employed by businesses and are mainly
aimed at sentiment analysis of customer feedback about goods and services. There is
an extensive selection of tools for textual analysis, including open-source tools and
multi-performance business toolkits and platforms (Batrinca & Treleaven, 2015).
This section introduces essential tools and toolkits to collect, clean, and analyze.
Popular scientific analytics tools, such as R, MATLAB®, Python, and Mathematica,
have been improved to support retrieving and analyzing text. Additionally, network
visualization tools, such as NodeXL and UCINET, enable researchers to draw rela-
tionships among actors or entities of social media within the network.

Scientific Programming Tools

R is a language and open-source software tool for statistical computing and graph-
ics, which provides a wide variety of statistical techniques, including linear and
nonlinear modeling, classical statistical tests, time series analysis, classification, and
clustering, and graphical methods (www.r-project.org). MATLAB® is designed for
numeric experimental programming, including data processing and data modeling,
which enables time series analysis, GUI and array-based statistics. MATLAB® can
212 ◾ Analytics and Knowledge Management

be employed for an extensive range of applications, and significantly improved pro-

cessing speed when compared to the traditional programming languages.
Python is a high-level programming language for general purpose program-
ming, and it has gained popularity because the language is comparatively easy
to learn and it enables programmers to write clear programs on both a small and
large scope. Python can deal with natural language detection, topic extraction, and
query matching. It also can be established to conduct sentiment analysis using dif-
ferent classifiers.
Apache Unstructured Information Management Applications (UIMA) is an
open-source project that analyzes “Big Data” and discovers information that is rel-
evant to the user. Mathematica is adopted for allegorical programming such as
computer algebra.

Network Visualization Tools

NodeXL is an add-in for Excel and supports the analysis and exploration of social
media with importing features that can pull data from Twitter, Flickr, YouTube,
Facebook, WWW hyperlinks, and wikis (Hansen, Shneiderman, & Smith, 2011).
NodeXL allows users to create visual network diagrams of actors (vertices), esti-
mate the network influence (e.g., betweenness centrality or page rank) of a single
actor on others, and retrieve information on a large scale (Hansen et al., 2011).
NodeXL has a professional version supporting more functions than its free version.
UCINET, developed by Lin Freeman, Martin Everett, and Steve Borgatti, is a
software package for the analysis of social network data. UCINET only runs on the
Windows operating system and comes with the NetDraw which is a social media
data visualization tool using a network format.

Business Applications
Business applications refer to commercial software tools that users can collect, search,
and analyze text for business purposes. For example, SAS Sentiment Analysis Manager
included in the SAS Text Analytics program allows users to access source content,
such as social media outlets, text data inside the organization, websites, and generate
reports about customers and competitors in real time (Batrinca & Treleaven, 2015).
RapidMiner is also a popular tool providing an open-source community edition as
well as a fee-based enterprise edition (Hirudkar & Sherekar, 2013). RapidMiner
offers data mining and ML procedures. These procedures include data extraction,
transformation, and loading (ETL), data visualization, structuring, assessment, and
employment. RapidMiner, written in Java, adopts learning schemes and attribute
evaluators using the Weka ML environment, and the R project is employed for sta-
tistical modeling schemes in RapidMiner. Similar to SAS Enterprise Miner, IBM
SPSS Modeler is one of the most popular applications that support various data
analytics tasks.
 Social Media Analytics ◾ 213

Social Media Monitoring Tools

In general, organizations adopt social media monitoring tools to conduct sentiment
analysis, and to trace and measure what users have been saying about a company,
products, services, and certain topics on social media. The examples of social media
monitoring tools contain Social Mention which offers social media notifications
called social mention alerts. Amplified Analytics emphasizes reviews regarding
products and services and marketing-related information. Reputation monitoring
tools include Lithium Social Media Monitoring and Trackur, which track conver-
sations and discussions online. Useful and free tools provided by Google include
Google Trends, which compares a specific search term or multiple search terms to
the total search volume, and Google Alerts, which is an automatic content notifica-
tion tool established around Google Search (Batrinca & Treleaven, 2015).

Text Analysis Tools

Text analysis tools are mainly designed for NLP and text analytics. Many open
source tools were developed by research groups and NGO (nongovernmental orga-
nizations) to capture, explore, and analyze opinions (Batrinca & Treleaven, 2015),
and examples include Stanford NLP group applications and LingPipe, a package
of Java libraries for the linguistic analysis of human language (Teufl, Payer, &
Lackner, 2010). Python natural language toolkit (NLTK), a popular open source
text analysis application, provides Python modules, linguistic data, and documen-
tation for text analytics. GATE, another example of open source sentiment analysis
tools, has been employed for media content analysis, customer experiences analysis,
cancer and pharmaceutical research, and web mining. Its website explains that sen-
timent analysis and polarity categorization are more challenging when compared to
traditional document classification.
Lexalytics Sentiment Toolkit is an automatic sentiment analysis applica-
tion. Though Lexalytics is not capable of data scraping, it is especially powerful
when applied to a large number of documents. Other commercial applications
for text mining include IBM LanguageWare, SPSS Text Analytics for Surveys,
STATISTICA Text Miner, Clarabridge, WordStat, AeroText, and Language
Computer Corporation (Batrinca & Treleaven, 2015).

Data Visualization Tools

Combining BI competences, data visualization tools enable diverse users to make
sense of Big Data. Through interactive user interfaces, users can conduct explor-
atory data analysis with visualizations on smartphones and tablet computers. These
tools assist users in recognizing patterns and relationships in data. Impromptu visu-
alization on the data can discover trends and outliers, which also can be conducted
on the framework of Big Data, including Apache Hadoop or Amazon Kinesis.
214 ◾ Analytics and Knowledge Management

Examples of data visualization tools include SAS Visual Analytics, Tableau, Qlik
Sense, Microsoft Power BI, and IBM Watson Analytics.

Social Media Management Tools

Social media management tools can be handy especially when users run an
agency or manage social media for other businesses. Examples of ways to man-
age social media include generating content, retargeting content, engaging with
users, and sharing schedules and related content. Social media management tools
can help to perform these tasks effectively and efficiently. Though many of the
tools share similar characteristics, users can choose the best ones that suit their
needs. Notable social media management tools include Hootsuite, Buffer, Agora
Plus, Sprout Social, CoSchedule, Everypost, Sendible, Social Pilot, MavSocial, and
Friends+Me.
Almost all social media management tools are commercial and charge fees for
use, and among those examples, Hootsuite is one of the most popular applications.
One of the most important features of Hootsuite is that it enables users to assist and
plan content, gauge their social ROI, and run advertisements in social media in an
all-in-one system. Hootsuite allows users to manage various accounts and keywords,
link with more than 35 social networks, and multischedule social media posts.

Representative Fields of Social Media Analytics

Social media reflects dynamic human information interaction, which creates new
opportunities to understand people, organizations, and society. More and more
researchers and business professionals are discovering innovative methods of auto-
matically gathering, incorporating, and investigating rich data on social media.
Therefore, documenting all the cases of social media applications in one section is
challenging. As Batrinca and Treleaven (2015) mentioned, three indicative fields
of SMA include business, bioscience, and social science. The early adopters of
SMA were retail industries and finance. Retail businesses employ social media
to reinforce their brand awareness, service enhancement for customers, product
improvement, marketing strategies, information procreation, and even fraud
detection.
In finance, social media is applied to evaluate market sentiment, and news data
is adopted for trading (Batrinca & Treleaven, 2015). For instance, Bollen, Mao,
and Zeng (2011) discovered with near 90% accuracy, by evaluating the sentiment
of a random sample of Twitter data, that the Dow Jones Industrial Average (DJIA)
prices are interrelated with the Twitter sentiment two or three days earlier. Wolfram
(2010) employed Twitter data to train a support vector regression (SVR) model in
forecasting values of each NASDAQ stock. The experiment resulted in revealing a
“considerable advantage” for predicting future values.
 Social Media Analytics ◾ 215

In marketing, the UGC of social media is used for detecting online and offline
opinions of customers. Anderson and Magruder (2012) applied a regression dis-
continuity approach to identifying a causal relationship between Yelp ratings and
dining reservations in restaurants. The study discovered that the more star ratings
the restaurants get, the more reservations they obtained, resulting in a complete
booking for the top-rated restaurants. Srinivasan, Rutz, and Pauwels (2016) exam-
ined the influence of a mix of marketing activities on users’ Internet activity and
sales. Applying a vector autoregressive (VAR) model, the researchers investigated
the relationship between price and marketing channels, TV advertisements, the
number of commercial search clicks and website visits, and Facebook activity on
sales. In addition, paid search was influenced by TV advertisements and the num-
ber of Facebook likes, while it affects marketing channels, website visits, likes on
Facebook, and sales (Moe, Netzer, & Schweidel, 2017).
In the field of biosciences, social media is used to quit smoking, deal with obe-
sity, and supervise diseases by gathering related data on peers for behavioral change
initiatives. Penn State University biologists (Salathe et al., 2012) created novel
applications and methods to trace the transmission of contagious diseases through
the data from news websites, blogs, and social media.
In computational social science, social media has been adopted to observe pub-
lic responses to political issues, public agenda, events, and leadership. As an illustra-
tion, Lerman, Gilder, Dredze, and Pereira (2008) applied computational linguistics
to forecast the news impact on the public regarding political candidates. Others
explored how Twitter is used within the election context to predict election results
(DiGrazia, McKelvey, Bollen, & Rojas, 2013) by discovering candidates’ patterns
of political practice (Bruns & Highfield, 2013). These applications focused more on
the behaviors of the candidates and paid less attention to the behavior of the public.
Social media provides a new channel for political candidates to get closer to the
public, and these public spheres also open communication channels for the online
audience to connect with each other and get involved with antagonistic politics.

Conclusions
Social media is a fundamentally measurable, information communication technol-
ogy that extends web-based communications. Web 2.0 is the second transformative
phase of the WWW and the social nature of Web 2.0 makes it possible to foster
social media platforms, such as Facebook, Twitter, LinkedIn, Reddit, YouTube, and
wikis. These social media platforms provide open environments for collective intel-
ligence that creates collaborative content, and this cogenerated content increases its
value with increased adoptions. The convenient access to APIs of the social media
platforms has resulted in an explosion of social data creation and the use of SMA.
Besides identifying definitions and processes of SMA, this chapter focused on
introducing major techniques and tools for SMA. Presented techniques include social
216 ◾ Analytics and Knowledge Management

media data scraping, sentiment analysis, topic modeling, visual analytics, and stream-
ing processing. Representative fields of social media analytics are business, bioscience,
and computational social science. One of the critical issues regarding SMA is that
social media platforms are increasingly limiting access to their data to make profits
from their content. Data scientists and researchers have to find ways to collect a large
scale of social media data for research purposes with reasonable costs. Otherwise,
computational social science could be the privilege of big organizations, resourceful
government agencies, and elite scholars that can afford costly social media data and
therefore the studies they conducted would hardly be evaluated or replicated.

References
Anaya, L. H. (2011). Comparing latent dirichlet allocation and latent semantic analysis as classifiers
(Doctoral dissertation). University of North Texas, Denton, TX.
Anderson, M., & Magruder, J. (2012). Learning from the crowd: Regression discontinuity
estimates of the effects of an online review database. The Economic Journal, 122(563),
957–989.
Armano, D. (2014, July 23). Six social media trends for 2010. Retrieved July 7, 2017, from
https://hbr.org/2009/11/six-social-media-trends
Batrinca, B., & Treleaven, P. C. (2015). Social media analytics: A survey of techniques,
tools and platforms. AI & Society, 30(1), 89–116.
Bengston, D. N., Fan, D. P., Reed, P., & Goldhor-Wilcock, A. (2009). Rapid issue tracking:
A method for taking the pulse of the public discussion of environmental policy.
Environmental Communication, 3(3), 367–385.
Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent dirichlet allocation. Journal of
Machine Learning Research, 3, 993–1022.
Bollen, J., Mao, H., & Zeng, X. (2011). Twitter mood predicts the stock market. Journal of
Computational Science, 2(1), 1–8.
Borders, B. (2009). A brief history of social media. Retrieved December 5, 2010, from
http://socialmediarockstar.com/history-of-social-media.
Botan, I., Derakhshan, R., Dindar, N., Haas, L., Miller, R. J., & Tatbul, N. (2010).
SECRET: A model for analysis of the execution semantics of stream processing sys-
tems. Proceedings of the VLDB Endowment, 3(1–2), 232–243.
boyd, D., & Ellison, N. (2010). Social network sites: Definition, history, and scholarship.
IEEE Engineering Management Review, 3(38), 16–31.
Brooker, P., Barnett, J., & Cribbin, T. (2016). Doing social media analytics. Big Data &
Society, 3(2), 2053951716658060.
Bruns, A., & Highfield, T. (2013). Political networks on twitter: Tweeting the queensland
state election. Information, Communication & Society, 16(5), 667–691.
Bruns, A., & Liang, Y. E. (2012). Tools and methods for capturing Twitter data during
natural disasters. First Monday, 17(4), 1–8.
Card, S. K., Mackinlay, J. D., & Shneiderman, B. (Eds.). (1999). Readings in information
visualization: Using vision to think. San Francisco, CA: Morgan Kaufmann.
Carraher, S. M., Parnell, J., & Spillan, J. (2009). Customer service-orientation of small
retail business owners in Austria, the Czech Republic, Hungary, Latvia, Slovakia, and
Slovenia. Baltic Journal of Management, 4(3), 251–268.
 Social Media Analytics ◾ 217

Chandramouli, B., Ali, M., Goldstein, J., Sezgin, B., & Raman, B. S. (2010). Data stream
management systems for computational finance. Computer, 43(12), 45–52.
Chung, W. (2016). Social media analytics: Security and privacy issues. Journal of Information
Privacy and Security, 12(3), 105–106.
Cohen, A. H. (2010, December 27). 10 social media 2010 highlights. Retrieved July 8, 2017,
from https://www.clickz.com/10-social-media-2010-highlights-data-included/53386/
Cook, K. A., & Thomas, J. J. (2005). Illuminating the path: The research and development
agenda for visual analytics. Los Alamitos, CA: IEEE Computer Society.
Davenport, T. H., & Court, D. B. (2014, November 5). Analytics 3.0. Retrieved August 26,
2017, from https://hbr.org/2013/12/analytics-30
Deerwester, S., Dumais, S. T., Furnas, G. W., Landauer, T. K., & Harshman, R. (1990).
Indexing by latent semantic analysis. Journal of the American Society for Information
Science, 41(6), 391.
Dergiades, T. (2012). Do investors’ sentiment dynamics affect stock returns? Evidence from
the US economy. Economics Letters, 116(3), 404–407.
Diakopoulos, N., Naaman, M., & Kivran-Swaine, F. (2010, October). Diamonds in the
rough: Social media visual analytics for journalistic inquiry. In Visual analytics science
and technology (VAST), 2010 IEEE symposium on (pp. 115–122). IEEE.
Diesner, J., Aleyasen, A., Chin, C., Mishra, S., Soltani, K., & Tao, L. (2015). ConText: Network
construction from texts [Software]. Retrieved from http://context.lis.illinois.edu/
DiGrazia, J., McKelvey, K., Bollen, J., & Rojas, F. (2013). More tweets, more votes: Social
media as a quantitative indicator of political behavior. PloS One, 8(11), e79449.
Edosomwan, S., Prakasan, S. K., Kouame, D., Watson, J., & Seymour, T. (2011). The his-
tory of social media and its impact on business. Journal of Applied Management and
Entrepreneurship, 16(3), 79.
Evangelopoulos, N., Zhang, X., & Prybutok, V. (2012). Latent semantic analysis: Five meth-
odological recommendations. European Journal of Information Systems, 21(1), 70–86.
Fan, W., & Gordon, M. D. (2014). The power of social media analytics. Communications
of the ACM, 57(6), 74–81.
Foltz, P. W., Kintsch, W., & Landauer, T. K. (1998). The measurement of textual coherence
with latent semantic analysis. Discourse Processes, 25(2–3), 285–307.
Gottfried, J., & Shearer, E. (2016, May 26). News use across social media platforms 2016.
Pew Research Center.
Greenwood, S., Perrin, A., & Duggan, M. (2016, November 11). Social media update 2016:
Facebook usage and engagement is on the rise, while adoption of other platforms
holds steady. Pew Research Center.
Grubmüller, V., GÖtsch, K., & Krieger, B. (2013). Social media analytics for future ori-
ented policy making. European Journal of Futures Research, 1(1), 20.
Hansen, D., Shneiderman, B., & Smith, M. A. (2011). Analyzing social media networks with
NodeXL: Insights from a connected world. Burlington, MA: Morgan Kaufmann.
Hassan, S., Sanger, J., & Pernul, G. (2014, January). SoDA: Dynamic visual analytics of big
social data. In Big data and smart computing (BIGCOMP), 2014 international confer-
ence on (pp. 183–188). Piscataway, NJ: IEEE.
Hebrail, G. (2008). Data stream management and mining. Mining Massive Data Sets for
Security, IOS Press, 89–102.
Hirudkar, A. M., & Sherekar, S. S. (2013). Comparative analysis of data mining tools and
techniques for evaluating performance of database system. International Journal of
Computational Science and Applications, 6(2), 232–237.
218 ◾ Analytics and Knowledge Management

Holsapple, C., Hsiao, S. H., & Pakath, R. (2014). Business social media analytics: Definition,
benefits, and challenges. Twentieth Americas Conference on Information Systems,
Savannah, GA.
Informs.org. UPS On-road integrated optimization and navigation (ORION) project.
Retrieved September 6, 2017, from https://www.informs.org/Impact/O.R.-Analytics-
Success-Stories/UPS-On-Road-Integrated-Optimization-and-Navigation-ORION-
Project
Jasra, M. (2010, November 24). The history of social media [Infographic]. Retrieved
December 4, 2010, from Web Analytics World: http://www.webanalyticsworld.
net/2010/11/history-of-social-media-infographic.html
Junco, R., Heibergert, G., & Loken, E. (2011). The effect of Twitter on college student
engagement and grades. Journal of Computer Assisted Learning, 27, 119–132.
Kaplan, A. M., & Haenlein, M. (2010). Users of the world, unite! The challenges and
opportunities of Social Media. Business Horizons, 53(1), 59–68.
Kevthefont (2010). Curse of the Nike advert-it was written in the future. Bukisa, p. 1.
Kohut, A., Keeter, S., Doherty, C., & Dimock, M. (2008). Social networking and online
videos take off: Internet’s broader role in campaign 2008. TPR Center, The PEW
Research Center.
Kontopoulos, E., Berberidis, C., Dergiades, T., & Bassiliades, N. (2013). Ontology-based sen-
timent analysis of Twitter posts. Expert Systems with Applications, 40(10), 4065–4074.
Kumar, A., & Sebastian, T. M. (2012). Sentiment analysis on Twitter. IJCSI International
Journal of Computer Science Issues, 9(4), 372.
Kurniawati, K., Shanks, G. G., & Bekmamedova, N. (2013). The business impact of social
media analytics. ECIS, 13, 13.
Lerman, K., Gilder, A., Dredze, M., & Pereira, F. (2008, August). Reading the markets:
Forecasting public opinion of political candidates by news analysis. Proceedings of the
22nd International Conference on Computational Linguistics, 1, 473–480.
Liu, Y., Huang, X., An, A., & Yu, X. (2007). ARSA: A sentiment-aware model for predict-
ing sales performance using blogs. In Proceedings of the 30th annual international
ACM SIGIR conference on research and development in information retrieval (pp. 607–
614). Amsterdam, the Netherlands, July 23–27.
McCallum, A. K. (2002). MALLET: A machine learning for language toolkit: Topic mod-
eling. Retrieved November 4, 2016, from http://mallet.cs.umass.edu/topics.php
Mejova, Y. (2009). Sentiment analysis: An overview. Comprehensive exam paper.
Retrieved February 3, 2010, from http://www.cs.uiowa.edu/~ymejova/publications/
CompsYelenaMejova.pdf
Moe, W. W., Netzer, O., & Schweidel, D. A. (2017). Social media analytics. In
B. Wierenga and R. van der Lans (Eds.), Handbook of marketing decision models
(pp. 483–504). Cham, Switzerland: Springer. https://www.springerprofessional.de/
en/handbook-of-marketing-decision-models/13301802
Obar, J. A., & Wildman, S. S. (2015). Social media definition and the governance challenge:
An introduction to the special issue. Telecommunications Policy, 39(9), 745–750.
Pang, B., & Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and
Trends® in Information Retrieval, 2(1–2), 1–135.
Park, S., Ko, M., Kim, J., Liu, Y., & Song, J. (2011). The politics of comments: Predicting
political orientation of news stories with commenters sentiment patterns.
In Proceedings of the ACM 2011 conference on computer supported cooperative work
(CSCW ’11) (pp. 113–122). New York, NY: ACM.
 Social Media Analytics ◾ 219

Peña-López, I. (2007). Participative web and user-created content: Web 2.0, Wikis, and social
networking. Paris: OECD. Retrieved October 24, 2007, from http://213.253.134.43/
oecd/pdfs/browseit/9307031E.pdf
Rimskii, V. (2011). The influence of the Internet on active social involvement and the for-
mation and development of identities. Russian Social Science Review, 52(1), 79–101.
Ritholz, B. (2010). History of social media. Retrieved December 5, 2010, from http://www.
ritholtz.com/blog/2010/12/history-of-social-media/
Ruggiero, A., & Vos, M. (2014). Social media monitoring for crisis communication: Process,
methods and trends in the scientific literature. Online Journal of Communication and
Media Technologies, 4(1), 105.
Salathe, M., Bengtsson, L., Bodnar, T. J., Brewer, D. D., Brownstein, J. S., Buckee, C., …
Vespignani, A. (2012). Digital epidemiology. PLoS Computational Biology, 8(7),
e1002616.
Sidorova, A., Evangelopoulos, N., Valacich, J.S., & Ramakrishnan, T. (2008). Uncovering
the intellectual core of the information systems discipline. MIS Quarterly, 32(3),
467–482, A1–A20.
Sinha, V., Subramanian, K. S., Bhattacharya, S., & Chaudhary, K. (2012). The contem-
porary framework on social media analytics as an emerging tool for behavior infor-
matics, HR analytics and business process. Management: Journal of Contemporary
Management Issues, 17(2), 65–84.
Srinivasan, S., Rutz, O. J., & Pauwels, K. (2016). Paths to and off purchase: Quantifying
the impact of traditional marketing and online consumer activity. Journal of the
Academy of Marketing Science, 44(4), 440–453.
Stavrakantonakis, I., Gagiu, A. E., Kasper, H., Toma, I., & Thalhammer, A. (2012). An
approach for evaluation of social media monitoring tools. Common Value Management,
52(1), 52–64.
Stieglitz, S., & Dang-Xuan, L. (2013). Social media and political communication: A social
media analytics framework. Social Network Analysis and Mining, 3(4), 1277–1291.
Teufl, P., Payer, U., & Lackner, G. (2010, September). From NLP (natural language pro-
cessing) to MLP (machine language processing). In International conference on mathe-
matical methods, models, and architectures for computer network security (pp. 256–269).
Berlin: Springer.
Thelwall, M., Buckley, K., & Paltoglou, G. (2011). Sentiment in Twitter events. Journal of
the American Society for Information Science and Technology, 62(2), 406–418.
Westerski, A. (2008). Sentiment analysis: Introduction and the state of the art overview
(pp. 211–218). Madrid, Spain: Universidad Politecnica de Madrid.
Wihbey, J. (2015). How does social media use influence political participation and civic
engagement? A meta-analysis. Journalist Resource. Retrieved February 17 2016.
Wittwer, M., Reinhold, O., & Alt, R. (2016). Social media analytics in social CRM-
towards a research agenda. In Bled eConference, p. 32.
Wolfram, M. S. A. (2010). Modelling the stock market using Twitter. Scotland, UK:
University of Edinburgh.
Zeng, D., Chen, H., Lusch, R., & Li, S. H. (2010). Social media analytics and intelligence.
IEEE Intelligent Systems, 25(6), 13–16.
Chapter 8

Transactional
Value Analytics
in Organizational
Development
Christian Stary

Contents
Introduction ..................................................................................................... 222
Value Network Analysis .....................................................................................223
Organizations as Self-Adapting Complex Systems .............................................224
Patterns of Interaction as Analytical Design Elements....................................... 226
Tangible and Intangible Transactions ................................................................ 226
Value Network Representation of Organizations ...............................................227
Analyzing the Value Network ............................................................................231
How to Ensure Coherent Value Creation ..........................................................238
Eliciting Methodological Knowledge .................................................................238
Externalizing Value Systems through Repertory Grids .......................................241
Developing Commitment for an Organizational Move .....................................245
Conclusive Summary ....................................................................................... 246
References .........................................................................................................247

221
222 ◾ Analytics and Knowledge Management

Introduction
Stakeholders are increasingly involved in organizational change and development,
and thus, value creation processes (Tantalo & Priem, 2016). However, stakeholders
whose values are incongruent with the values of their organization are likely to be
disengaged in work (Rich, Lepine, & Crawford, 2010), and finally, organizational
development processes (Vogel, Rodell, & Lynch, 2016). Understanding value con-
gruence, that is, the relationship between stakeholder values and values created on
the organizational level, is therefore key to developing organizations (Edwards &
Cable, 2009), and a prerequisite to developing a culture for high performance
(Posner & Schmidt, 1993).
To avoid the emergence of behavior (patterns), for example, pretending to fit
into an organization of work (Hewlin, Dumas, & Burnett, 2017), when stake-
holders feel that their values do not match those of the organization, a deeper
analysis of values of self-knowledge can help. It allows addressing relation-
ships between value congruence, perceived organizational support, core self-
evaluation, task performance, and organizational citizenship behavior (Strube,
2012). In the following, self-knowledge analytics based on individual stakeholder
transactions as an integral part of an enriched value network analysis (VNA)
approach (Allee, 2008), is introduced. Thereby, interactions between stakeholder
roles relevant for operation are analyzed. They are represented in diagrammatic
networks (termed holomaps) and can be tangible or intangible. Tangible interac-
tions encode role-specific deliverables that are created by persons acting in a cer-
tain role and need to be exchanged for task completion. Intangible interactions
encode deliverables facilitating business operations that are not part of formal role
specifications. However, they are likely to influence the successful completion of
business operations.
Since VNA targets organizational change as considered possible by stake-
holders, individual value systems need to be analyzed before changes can become
effective. This organizational transaction analysis ensures interactional coherence
based on individual value systems. In business-to-customer relationships, value
cocreation is well established (Payne, Storbacka, & Frow, 2008; Vargo, Maglio, &
Akaka, 2008). The subjective analytics in value alignment help to identify indi-
vidual capabilities and needs, which otherwise might be overlooked or unrecog-
nized (Jankowicz, 2001). In large organizations, software repositories may help
in understanding and evaluating requirements and encoding needs for change
(Selby, 2009). Dashboards have been designed not only to display complex infor-
mation and analytics but also to keep indicators with respect to volatility. The latter
requires stakeholder judgement, also termed as a user- or demand-driven approach
in analytics (Lau, Yang-Turner, & Karacapilidis, 2014), which focuses on participa-
tive elicitation processes, recognizing that any data analysis needs to make sense for
the involved stakeholders.
 Transactional Value Analytics in Organizational Development ◾ 223

Studies in that context reveal that decision making on variants or proposals not
only requires individual intellectual capabilities but also collective knowledge and
active sharing of expertise (Lau et al., 2014). In particular, in complex work settings,
people with diverse expert knowledge need to work together toward a meaning-
ful interpretation of interactional change (Treem & Leonardi, 2017). Thus, any
consensus-building mechanism needs to accommodate input from multiple human
experts effectively. This work proposes to enrich VNA methodologically by using
repertory grids (Easterby-Smith, 1976; Fransella & Bannister, 1977; Senior, 1996;
Boyle, 2005) to embody individual value systems into organizationally relevant
ones. This enrichment is intended to facilitate decision making in organizational
change management processes.
The chapter starts with an introduction to the methodological challenge by
providing the relevant background on VNA. It identifies those elements in VNA
that may cause fragmentation of value-driven alignment of change proposals,
which should become part of a more structured consensus-building process, in par-
ticular when business processes should be developed on that ground (Stary, 2014).
Several interviews have been conducted with experts dealing with aligning value
systems in stakeholder-driven change management, to identify suitable candidates
to enable value-system-based consensus building. A report on the interviews and
the evaluated candidates is given. Based on the results, the consensus-building
process can be supported by individually-generated repertory grids, which are
introduced subsequently. They allow individuals in certain roles to reflect on
their value system when acting in that role. In this way, individual stakeholders
can rethink organizational behavior changes offered by or to other stakeholders
and assess them according to their role-specific value system. An exemplary case
demonstrates the feasibility of this methodological enrichment and reveals first
insights from a typical use scenario, namely organizational sales development.
We conclude the chapter by summarizing the achievements and providing topics
for further studies.

Value Network Analysis

Since its introduction, VNA has targeted a key question of business, namely “How
is value created through converting intangible assets to tangible ones, and thus
negotiable forms of value?” (Allee, 2008). The method aims at developing orga-
nizations beyond the value chain, since traditional value-chain models represent
a linear, if not mechanistic, view of business and its operation. Complex constel-
lations of values, however, require analyzing business relationships by taking into
account the role of intangible value exchange as a foundation for value creation.
Value exchange needs to be analyzed before changing business transactions in prac-
tice. In particular, complex relationships require preprocessing from a value-based
224 ◾ Analytics and Knowledge Management

perspective, as they influence effectiveness and efficiency, and cause possible fric-
tion in operational processes (Allee, 2008).
VNA is meant to be a development instrument beyond engineering, as it aims
to understand organizational dynamics, and thus to manage structural know-
ledge from a value-seeking perspective, for individuals and the organization as a
whole. However, it is based on several fundamental principles and assumptions
(Allee, 1997, 2003, 2008)1:

◾ Participants of an organization and organizationally-relevant stakeholders

participate in a value network by converting what they know, both individu-
ally and collectively, into tangible and intangible values that they contribute
to the network, and thus to the organization.
◾ Participants accrue value from their participation by converting value inputs
into positive increases of their tangible and intangible assets, in ways that will
allow them to continue producing value outputs in the future.
◾ In such a network, each participant contributes and receives value in ways
that sustain both their own success and the success of the value network as a
whole. This mutual dependency is a condition sine qua non. Once active par-
ticipants either withdraw or are expelled, the overall system becomes unstable
and may collapse, and needs to be reconfigured.
◾ Value networks require trusting relationships and a high level of integrity and
transparency on the part of all participants. Then, insights can be gained into
interactions by identifying and analyzing not only the patterns of exchange,
but rather the impact of value transactions, exchanges, and flows, and thus
the dynamics of creating and leveraging value.
◾ A single transaction is only meaningful in relation to the system as a whole. It
is set by role carriers who utilize incoming deliverables from other role carriers
(inputs) and can assess their value, and they realize value that is manifest by
generating output.

As stakeholders—in roles relevant for business—are responsible for their relations

with others, the organization itself needs to be conceptualized as a highly dynamic
and complex setting.

Organizations as Self-Adapting Complex Systems

Organizations as sociotechnical systems are complex living systems and as such
are self-regulating and self-managing (Allee, 2003). When managing organiza-
tions from the outside in, by predetermining structures, systems, rules, and

1 For the sake of preciseness, the VNA description closely follows the original texts provided by
Allee (2003).
 Transactional Value Analytics in Organizational Development ◾ 225

formal reporting relationships, inherent dynamics may get overlayed and become
completely controlled externally. In rapidly changing settings, self-organization
of concerned stakeholders is an effective way to handle requirements for change.
However, for self-organization to happen, stakeholders need to have access to
relevant information and, more importantly, an understanding of the organization
and the situation as a whole. Both are required to make informed decisions and
initiate socially effective action. Since the behavior of autonomous stakeholders
cannot be predicted fully, organizations need rules to guide behavior management
according to the understanding of stakeholders and their capabilities to change
their behavior.
This need can be adeptly demonstrated in customer service. When there are
too many rules, customers are locked into a bureaucracy that seems unresponsive
to their needs. When there are too few rules, inconsistency and chaos are likely
in complex business cases. Stakeholders need to develop guiding principles that
effectively support them in organizational design and the respective decision mak-
ing through information and technology provision (Firestone & McElroy, 2003).
These principles need to tackle both tangible and intangible relationships and
stakeholder interaction. They should be qualified to reflect on their tangible and
intangible exchanges, and finally negotiate their own “protocols,” that is, activities
with those with whom they interact (Allee, 2008).
Although no one person or group of people can manage a complex system, the
stakeholders can self-organize their inputs and outputs and negotiate exchanges
with others in the organizational system as necessary. Modeling work and busi-
ness relations as dynamic patterns of tangible and intangible exchanges help
stakeholders to identify individually consistent roles and understand the system.
They also allow them to make it transparent and therefore communicable with
coworkers, business partners, and the economic ecosystems of which they are
part. According to Allee (2003), all organizational layers are concerned with the
following:

◾ Operationally: Stakeholders need to understand how digital networks and

technologies support creating, organizing, and accessing the everyday know-
ledge they need to complete their tasks and make informed decisions.
◾ Tactically: They need to understand how social webs such as knowledge net-
works and communities of practice help create, diffuse, and leverage know-
ledge and innovation.
◾ Strategically: Stakeholders need to develop an understanding of their organi-
zation as a participant in (multiple) business networks, where intangibles are
important for building relationships and work interactions.

Above all, stakeholders need to accept the dual nature in interaction in networked
ecosystems through tangibles and intangibles in order to learn how to engage in
conversations that matter.
226 ◾ Analytics and Knowledge Management

Patterns of Interaction as Analytical Design Elements

VNA builds upon organizations as self-adapting complex systems, which are mod-
eled from that perspective by the following:

1. Identifying patterns of interactions representing tangible and intangible rela-

tions between stakeholder roles.
2. Describing these patterns in a structured way, recognizing:
a. Sources and sinks of information exchanged between stakeholder roles.
b. The impact of received information and objects.
c. The capabilities of produced and delivered assets.
3. Elaborating critical processes or exchanges, and thus proposing changes from
both a cognitive perspective and the flow of energy and matter.

In line with the living system perspective, VNA assumes that the basic pattern of
organizing a business is that of a network of tangible and intangibles exchanges.
Tangible exchanges correspond to flows of energy and matter, whereas intangible
exchanges point to cognitive processes. Describing a specific set of participating
stakeholders and exchanges allows a detailed description of the structure of any
specific organization or a network of organizations.
Although VNA considers the act of exchange to be a fundamental activity,
it goes beyond traditional economic understanding of stakeholder interactions.
Exchange includes goods, services, and revenue, but also considers the transaction
between stakeholders as representative of organizational intelligence, thus as a cog-
nitive interaction process. Transactions ensure successful task accomplishment and
business through cognitively reflected exchanges of information and knowledge
sharing, opening pathways for informed decision making. Hence, exchanges not
only have value per se, but encode the currently available collective intelligence
(determining the current economic success).

Tangible and Intangible Transactions

In VNA, knowledge, either implicit or explicit (Dalkir, 2011), and intangible
exchanges are different from tangible ones, and as such they need to be specifi-
cally treated in relation to their characteristics. Tangible exchanges include goods,
services, and revenue, in particular physical objects, contracts, invoices, return
receipts of orders, requests for proposals, confirmations, and payments. They also
include knowledge, products, or services that directly generate revenue, or that are
expected (contractual) and paid for as part of a service or good.
Intangible exchanges comprise knowledge and benefits. Intangible knowledge
and information exchanges occur and support the core product and service value
chain, but are not contractual. Intangibles are extras that stakeholders in a certain
 Transactional Value Analytics in Organizational Development ◾ 227

role provide to others to help keep business operations running. For instance, a
service organization asks sales experts to volunteer time and knowledge on organi-
zational development in exchange for an intangible benefit of prestige by affiliation.
Stakeholders involved in intangible transactions help to build relationships by
exchanging strategic information, planning knowledge, process knowledge, and
technical know-how, and in this way they share collaborative design work, perform-
ing joint planning activities and contributing to policy development. Intangibles,
like other assets, are increased and leveraged through deliberate actions. They affect
business relationships, human competence, internal structure, and social culture.
VNA considers intangibles as assets and negotiables that can actually be delivered
by stakeholders engaged in a knowledge exchange. They can be held accountable
for the effective execution of that exchange, as they are able to articulate them
accordingly when following the VNA’s structured procedure.
Although there are various attempts to develop new measures and analytical
approaches for calculating knowledge assets and for understanding intangible value
creation, traditional scorecards need to move beyond considering people as liabilities,
resources, or investments. Responsible stakeholders need to understand how intan-
gibles create value and, most importantly, how intangibles go to market as negotiables
in economic exchanges. As a prerequisite, they need to understand how intangibles
act as deliverables in key transactions with respect to a given business model.

Value Network Representation of Organizations

Value networks represent organizations as a web of relationships that generates tan-
gible and intangible values through transactions between two or more roles. These
roles stem from any public or private organization or sector and stand for individu-
als, groups, or entire organizations. The network, instead of representing hierarchical
positions, structures the dynamics of processing and delivering tangibles and intan-
gibles. Although the roles need to be related to the organization at hand, suppliers,
partners, and consumers—regardless of their physical location—need to become part
of the network once they generate value or receive transactional deliverables.
When modeling an organization as a value network, several assumptions apply
(Allee, 2008):

◾ An exchange of value is supported by some mechanism or medium that

enables the transaction to happen. As organizations can also be considered
sociotechnical systems, typical enablers are information and communication
technologies. For instance, a sales briefing is scheduled by utilizing some spe-
cific web application, such as doodle.com.
◾ There is provided value, which in relation to the briefing is based on a tangible
exchange of inputs of customer service, and response to inquiries between
organizers and participants. The intangibles are targeted news and offerings
228 ◾ Analytics and Knowledge Management

as well as updates on services and customer status (knowledge), and a sense

of community (benefit).
◾ There is return value, which is efficiency in terms of short handling time of
customer requests as tangible, and informed customer request and feedback on
latest developments (knowledge) and customer loyalty (benefits) as intangibles.

Value exchanges are modeled in a special type of concept map (Novak & Cañas,
2006), termed a holomap. Concept maps have turned out an effective means to
articulate and represent knowledge (cf. Trochim & McLindon, 2017). They have
been in use since the 1980s for acquiring mental models while graphically gener-
ating a coherent conceptual model, both supporting individuals, and groups in
sharing and planning processes (cf. Goldman & Kane, 2014). Recently, concept
maps served as a baseline for value analytics (cf. Ferretti, 2016), and have been used
effectively for congruence analysis of individual cognitive styles (Stoyanov et al.,
2017). Their ease of use allows triggering learning processes in distributed settings,
for example, supported by web technologies (cf. Wang et al., 2017).
The VNA mapping from the observed reality to a role-specific concept map
(holomap) is based on the following elements:

◾ Ovals represent the functional roles of stakeholders, termed participants of

the value network, that is, the nodes of the network.
◾ Participants send or extend deliverables to other participants. One-directional
arrows represent the direction in which the deliverables are moving during a
specific transaction. The label on the arrow denotes the deliverable.

When modelers create holomaps, they think of participants as persons they know
carrying out one or more roles in the organizational system at hand. Holomapping
is based on the assumption that only individuals or groups of people have the power
to initiate action, engage in interactions, add value, and make decisions. Hence,
VNA participants can be individuals, small groups or teams, business units, whole
organizations, collectives such as business networks or industry sectors, communi-
ties, or even nation-states. VNA does not consider databases, software, or other
technology to be a participant. It is the decision-making capability about which
activities to engage in that qualifies only humans as VNA participants.
Transactions or activities are represented by an arrow that originates with one
participant and ends with another. The arrow represents movement and denotes the
direction of addressing a participant. In contrast with participants, which tend to
be stable over time, transactions are temporary and transitory in nature. They have
a beginning, a middle, and an end point.
Deliverables are those entities that move from one participant to another. A deliv-
erable can be physical or tangible, like a document or a physical object, but it can also
be nonphysical, such as a message or request that may only be delivered verbally. It
can also be an intangible deliverable of knowledge about something, or a favor.
 Transactional Value Analytics in Organizational Development ◾ 229

In VNA, an exchange only occurs when a transaction results in a particular

deliverable coming back. A gap is when something is provided without anything
being received in return. However, focusing on the exchange as the molecular
element of value creation is a generic concept that enables capturing a variety of
organizations as value networks. Tangible and intangible exchanges establish pat-
terns typical of business relationships. In many cases, tangible exchanges comprise
exchanges of matter and energy (goods and money), while the intangible exchanges
capture cognitive and emotive exchanges such as favors and benefits.
In the following, we provide an example of a VNA case in the Sales and Presales
group of a service company providing innovative instruments (methods and tech-
nologies) for knowledge acquisition and sharing. To understand the overall patterns
of exchange, and to determine the impact of tangible and intangible inputs for each
participant, the company decided to perform a VNA. This should not only help to
analyze the state of affairs, but also leverage potential changes for each participant.
The Sales and Presales group aims to improve their ability to utilize operations and
customer feedback in further developing their services.
The first step participants need to consider in the modeling process involves all
the roles, organizational units, or work groups, both internally and externally, that are
considered of relevance in the activities of the Sales and Presales group. In this case,
four groups (participants) inside the organization are identified, namely Sales and
Presales, Product Development, Customer Service, and freelance interviewers outside
the organization. They represent the nodes in the holomap in Figure 8.1.
For modeling, first the stakeholders need to think about tangible exchanges
that take place between the participants. What are the transactions adding value?
What are the tangible deliverables in the work system? Figure 8.1 shows tangible
deliverables such as product information, feedback from market, requests, and
updates. For these cases, the transaction and communication channel is considered
a tangible deliverable because it either comprises core data relevant for operating
the business, or affects essential relations to organizational units for product and
(customer) knowledge management.
Intangible transactions or exchanges are modeled the same way. To distinguish
the intangible deliverables from the tangible deliverables, modelers use a different
line style (dotted line in Figure 8.1). For the service provider at hand, intangibles
are incomplete information, involving an order handling report, customer report,
customer preparation data, and so on, which various stakeholders make available
through reporting and active sharing of knowledge (Figure 8.1). They are consid-
ered intangible because there is no direct monetary income related to them. They
are neither contracted by the provider nor expected by the recipients. They are extra
offerings to participants to keep the operation running, and product development
informed, mainly based on informal learnings and experiential knowledge.
As shown in Figure 8.1, several tangible exchanges occur. For example, the
Product Development group provides product announcements to the Presales group
in exchange for feedback from the market, whereas the Sales group provides requests
 Feedback to product

Product information

Order handling Updates

Customer report Product
service development
Incomplete
Acknowledgement information
Customer Update
information Interview Order request Feedback
data Sales from market
Leads
Request for Product
clarification Request announcement
Delayed
delivery
Acknowledgement
Interviewer Presales
Incomplete
230 ◾ Analytics and Knowledge Management

information

Customer Customer
report preparation data

Quality report

Figure 8.1 Sample holomap for developing sales and presales.

 Transactional Value Analytics in Organizational Development ◾ 231

to interviewers who acknowledge them. In addition, several intangible exchanges

occur, such as interviewers provide customer information to the Customer Service
group in exchange for customer reports. The latter complements the formal, role-
specific exchange specified through the pair “request for clarification—interview
data.” It documents the intention to provide a comprehensive picture of customers to
build trustful relationships with customers (representing the benefits of the exchange).
However, several one-sided transactions, with respect to tangibles and intangibles,
become evident, as also shown in the holomap in Figure 8.1. For instance, concerning
intangibles, the interviewers provide customer preparation data and quality reports
to the Presales and Product Development groups respectively, without any intangible
return. Concerning tangibles, the Product Development group provides both prod-
uct information and updates to the Sales group without any return, for example.
Once all exchanges and deliverables are captured in the holomap, a dia-
gram of how the business is perceived from a stakeholder perspective is estab-
lished. The value network view of an organization helps understand the role
of knowledge and intangibles in value creation. The modeling process allows
capturing strategically critical intangible exchanges from a stakeholder perspec-
tive, thus enabling further targeting opportunities for value creation. This issue
is addressed through analyzing the value network as represented by the holomap
in the next subsection.

Analyzing the Value Network

According to Allee (2003, 2008), analyzing the capabilities of a value network and
developing opportunities for constructive change requires three different types of
analysis. Transactional value analytics is based on elicited transactions between people
acting in certain roles and explores patterns in terms of reciprocity and the perceived
value of generated deliverables transmitted to other stakeholders, and received deliver-
ables from other stakeholders. Based on this data, change proposals can be developed
in terms of offering novel deliverables to other stakeholders, either changing their type
from intangible to tangible, generating novel ones, or including further stakeholders
not involved in interaction from the perspective of a certain role carrier so far.
The first analysis is about assessing the structure and dynamics of the repre-
sented system as a whole. The second and third analyses focus on each participant’s
role in the interactions of the value network:

1. Exchange analysis: What is the overall pattern of exchanges in the represented

system?
2. Impact analysis: What impact does each value input have on the participants?
3. Value creation analysis: How can an organization create, extend, and leverage
value, either through adding value by participants, extending value to other
participants, or converting one type of value to another?
232 ◾ Analytics and Knowledge Management

The exchange analysis investigates the overall patterns of exchanges addressing a

variety of structural issues. When starting to identify missing relations for operat-
ing the business or links requiring a rationale, potential breakdowns in flow that
can be critical for the business can be determined. In that context, the coherence of
relations and resulting flows of how value is generated in the network can be evalu-
ated. By checking the amount of tangibles and intangibles, a possible dominance
of a particular type of exchange can be determined. The overall pattern of reci-
procity reveals involvement data of the participants (as perceived by the respective
modeler). Extensive sources and sinks of interactions should be noted as potential
for optimizing the entire network, avoiding specific participants benefiting at the
expense of others.
In the holomap in Figure 8.1, several patterns can be noticed. The Sales and
Presales groups gain knowledge about the product from the Product Development
group, but that knowledge results in dead ends and so is not passed on explicitly to
the Customer Service group and interviewers. In fact, there is no significant prod-
uct knowledge exchange between Sales, Presales, Customer Service, and interview-
ers. Another pattern that shows up is that knowledge from the Customer Service
group to the Product Development group only flows one way. The Customer
Service group delivers feedback on a product to the Product Development group
as intangible, but does not have any channel for a two-way communication about
product-related data with the Product Development group.
Allee (2003, 2008) states there would be no straightforward interpretation for
what certain patterns mean, although they become apparent through the exchange
analysis. Patterns trigger discussions among those participating in identified pat-
terns, in particular when missing steps in a core process or cultural issues regarding
the organization pop up. According to practical experience, it is the set of upcom-
ing questions that lead to informed impact and value creation analysis, and thus
change.
The impact analysis focuses on received inputs that trigger some response by the
receiver. The analysis considers activities and effort for handling the input and for
leveraging the value received. The concerned stakeholders need to estimate the tan-
gible and intangible costs (or risks) and gains for each input. They need to describe
how a certain activity is generated in response to a certain (tangible or intangible)
input. Then, they are asked to describe the estimated increase or decrease of tan-
gible assets in terms of overall cost or risk-benefit analysis (Table 8.1).
Table 8.1 shows the value inputs for the Sales group. Each of the received inputs
is explored according to the various activities to be set, as well as the costs and
benefits it brings. In this way, each Sales participant can address the value impact
for the whole organization. Each input presented in the holomap (Figure 8.1) is
described using a single line as it may affect essential tangible and intangible inputs
to the Sales and Presales groups. For instance, the fourth-line item shows incom-
plete delivery of information from the Presales group to the Sales group. This results
in a low value since the required knowledge on sales transactions was not available
Table 8.1 Sample Impact Analysis of a Sales Participant
Transactions Impact Analysis

Impact on
Financial Perceived
Resources Overall Cost/ Value in
Activities That are (Positive/ Risk for This Overall Benefit of View of
Deliverable From To Generated Negative) Impact on Intangible Assets (Positive/Negative) Input This Input Recipient

Human Internal Business H = High H = High +2, +1

competence structure relationships M = Medium M = Medium Neutral
L = Low L = Low −1, −2

Order handling Customer Sales Evaluation of Extra time and Knowledge on Contact to Knowledge on H: 2 hours/ M: Neutral
report (Intangible) service report effort to be calculation presales competition M: recalcula- Documentation
calculated schema and and market tion required of order
accounting handling

Incomplete Product Sales Additional Effort to be Knowledge on Expert Technical skills H: 3 hours/ M: Completeness Neutral
information development information spent for product interviews and H: Availability of product
(Intangible) from product information knowledge of experts information
development to collection required
be collected

Delayed delivery Presales Sales Delay in Loss of service – Presales Presales H: Extension of L −2
(Tangible) customer time contact reminder loss of service
service to be time/
communicated H: Loss of
loyalty

Incomplete Presales Sales Additional Effort to be Knowledge on Presales Appointment M: 1 hour/ M: Completeness −1
information information spent for sales contact M: Availability of product
(Intangible) from presales to information transaction of presales information
be collected collection required
Transactional Value Analytics in Organizational Development ◾

(Continued)
233
 234 ◾

Table 8.1 (Continued) Sample Impact Analysis of a Sales Participant

Transactions Impact Analysis

Impact on
Financial Perceived
Resources Overall Cost/ Value in
Activities That are (Positive/ Risk for This Overall Benefit of View of
Deliverable From To Generated Negative) Impact on Intangible Assets (Positive/Negative) Input This Input Recipient

Acknowledgement Customer Sales Recognition — Customer Order Customer — H: Order +2

(Tangible) service service process contacted processed
satisfied

Acknowledgement Interviewer Sales Recognition — Interviews Order Customer — H: Order +2

(Tangible) enabled process contacted processed

Product information Product Sales Material to be Time and Knowledge Informed Qualification H: 2 hours per H: Informed +1
(Tangible) development studied effort to be acquisition external feature/ collaboration
studied contacts M: Request for
clarification
Analytics and Knowledge Management

Updates (Tangible) Product Sales Feature(s) to be Time and Knowledge Informed Qualification H: 1 hour per H: Informed +2
development studied effort for acquisition method feature/ collaboration
studying application M: Request for
clarification

Leads (Tangible) Presales Sales Update of Time and Knowledge on CRM- Data accuracy M: 5 min per H: Documentation +2
CRM-database effort for opportunities timeliness record/ of opportunities
update M: requires
clarification
 Transactional Value Analytics in Organizational Development ◾ 235

in time. The Presales group needs to be contacted for availability to overcome short-
comings in customer relationship management.
The value creation analysis not only reflects the situation as it is but supports the
proposal of changes to the way a participant is committed to the delivery. Although
the structure of this analysis is similar to the impact analysis, it instead focuses
on a participant’s capability regarding how to extend value to other participants
represented in the holomap. It analyzes the tangible and intangible costs (or risks)
and gains for each value output. Each value output could add new tangible or
intangible value and thus extend value to other participants represented in the holo-
map. By assessing each value output, a participant needs to determine the activities,
resources, and processes required, as well as the costs and anticipated benefits of
each value-creating activity.
In Table 8.2 a sample analysis is given for the Sales group. An important value
creation for the Sales group is to add value to product features received from the
Product Development group by providing timely information for interviewers and
the Customer Service group (in terms of what is actually ordered). Their role in
contributing to timely information in cooperation with the Product Development
group is actually very small, being of both low cost for them and also low benefit
in case of not actively requesting updates. Hence, they could be an active agent
to extend value through targeted requests and other efforts to reach the Product
Development group. They would need to engage in a value conversion process in
which they convert one relevant type of value to another with respect to a work-
ing product lifecycle. Engaging in increase value creation actually explores new
strategic possibilities and changes costs and benefits for the organization. It is the
moment where a member of the organization develops an offer to others by indi-
vidual behavior change instead of elaborating the expected behavior of others.
Looking at the value outputs that go to the Product Development group, the
Sales group can open a tangible communication channel to support the process
used by the Product Development group, and an intangible transaction of feeding
back product-relevant information. In this way, the Sales group can leverage their
intangible value outputs (information about product use) into more advanced prod-
uct development that could be turned into a revenue stream. In the example, the
communication channels were mostly one way, conveying product knowledge from
the Product Development group to the Sales group. With a focus on converting
that expected communication channel with another type of value gain, the Sales
group enables communication after gaining feedback for the Product Development
group. Both converting an intangible value to gain a tangible value, and establish-
ing an intangible deliverable with product feedback, supports the strategic intent of
a rapid response to the changing needs of customers.
Although value creation analysis can become very rich due to anticipated
changes in the flow of deliverables, the participants need to understand what
impact a particular output has on the participant who receives it. From an organi-
zational development perspective, a participant could maximize the effectiveness
 236 ◾

Table 8.2 Sample Value Creation Analysis

Transactions Perceived Value Value Creation Analysis

Recipient
Highly Values
This
Deliverable.
Strongly How High Is
Agree(+2) Tangible The Risk Factor
Agree(+1) Asset What Are The In Providing
Neutral(0) Utilization Is: Tangible Costs? This Output?
Disagree(−1) H = High (Financial And H = High
Strongly M = Medium Physical M = Medium
Deliverables From To Disagree(−2) L = Low Resources) L = Low

Update request Sales Product −1 L Processing time L

Analytics and Knowledge Management

(Intangible) development by request

Request Sales Interviewer +2 H Interview M

(Tangible)

Order Sales Customer service +2 H Processing time M

(Tangible) per order

(Continued)
Table 8.2 (Continued) Sample Value Creation Analysis
Value Creation Analysis (Continued)

Intangible asset
utilization is:
H = high
M = medium
L = low
(Human How do we What is the What is the
competence add to, overall overall
Internal structures What are other intangible costs or enhance, combined benefit for us
Business benefits? or extend cost/risk for in providing
relationships) (Industry, society, environment) value? this input? this input?

HC IS BR Industry Society Env

M M Benefit-timely — — Formal Formulating Trigger to

information procedure request/— product
M development

H H Customer — — — — High
care customer
H satisfaction

H H Order — — CRM- Delivery of Enriched

management Database update/ CRM-
update completeness Database
Transactional Value Analytics in Organizational Development ◾

H of data
237
238 ◾ Analytics and Knowledge Management

and efficiency of a certain business operation when following the created value.
The overall cost benefit analysis could result in excellent data. However, a closer
analysis of the proposed value creation could lead to inconveniences of the involved
participants, in particular business partners and customers (Augl & Stary, 2015,
2017). For instance, collecting context data from interviewers when applying the
product in the field could easily lead to rejecting the proposed value creation once
they consider a request for preparing such a report as a negative value input.
Before proposed value creations can become effective in business operations,
they need to be acknowledged by the involved participants as well as by respon-
sible management. To allow the constructive elaboration of value creations affect-
ing the collective, each member of the organization should be empowered with an
instrument enabling him or her to reflect on individual values. Applying such an
instrument requires stepping out of the VNA logic while providing a baseline for
discussing proposals for change. Once value-creation proposals are congruent with
individual value systems, the resulting business operation becomes coherent for the
involved stakeholders.

How to Ensure Coherent Value Creation

To facilitate the alignment of proposed value creations with individual value sys-
tems, repertory grids are suggested in this section. They allow inconsistencies of
value creations with individual value systems to be overcome, and are selected based
on several expert semi-structured interviews addressing value management from a
stakeholder perspective. Accordingly, in the first subsection the identification of a
proper method is described, before the repertory grids are detailed and exemplified
to demonstrate their analytical support capability in the context of the VNA.

Eliciting Methodological Knowledge

In search of a method allowing individual value systems to be externalized, 14 orga-
nizational analysts involved in value management were invited to a semi-structured
interview. The overarching question they were asked was “How far can organiza-
tional stakeholders be supported with methods when needing to align individual
with collective values or value systems?”
Each of the 14 analysts is active in Germany and Austria and have the follow-
ing profile:

◾ Long-term experience in practical value management, particularly in method-

driven approaches (e.g., utilizing Bohm’s Dialogue)
◾ Interest in theory-driven work praxis (e.g., system thinking)
◾ Experience in international organizational development projects
 Transactional Value Analytics in Organizational Development ◾ 239

The analysts were interviewed using the following sequence of items:

1. What is your particular work background with respect to value management?

2. What are your qualifications and background in organizational design, learn-
ing, and development by means of knowledge management techniques?
3. Which methods do you actually use with respect to value management?
4. Which methods do stakeholders need most when adjusting individual with
collective values or value systems?
5. Which method do you consider most effective when adjusting individual
value systems with collective ones? Kindly refer to:
a. Repertory grid
b. Critical incident technique
c. Storytelling
d. Another method you consider most effective based on your experience

The methods were picked from a selected collection of practically tested methods for
gaining and analyzing stakeholder knowledge (Stary, Maroscher, & Stary, 2012). The
repertory grid technique (Fransella & Bannister, 1977; Fromm, 1995; Goffin, 2002)
supports an understanding of how individuals perceive a certain phenomenon in a
structured way, namely by identifying the most relevant characteristics of the carriers
of those characteristics (e.g., roles, persons, objects, situations) on an individual scale.
The technique enables not only identifying attributes that relate to the carriers but
also to elicit explanations of these characteristics. Repertory grids have been intro-
duced by Kelly (1955) based on the personal construct theory in psychology, and
have been used successfully in a variety of contexts including expert, business, and
work knowledge elicitation (Ford, Perty, Adams-Webber, & Chang, 1991; Gaines &
Shaw, 1992; Hemmecke & Stary, 2006; Stewart, Stewart, & Fonda, 1981), strategic
management (Hodgkinson, Wright, & Paroutis, 2016), supplier-manufacturer rela-
tionship management (Goffin, Lemke, & Szwejczewski, 2006), learning support
(Stary, 2007), attitude studies (Honey, 1979), team performance (Senior, 1996;
Senior & Swailes, 2004), project management (Song & Gale, 2008), and consumer
behavior (Kawaf & Tagg, 2017).
The critical incident technique (Flanagan, 1954) is an exploratory qualitative
method that has been shown to be both reliable and valid in generating a compre-
hensive and detailed description of a content domain. The technique consists of
asking eyewitnesses for factual accounts of behaviors. The critical incident tech-
nique consists of a set of procedures for collecting direct observations of human
behavior in such a way as to facilitate their potential usefulness in solving practical
problems and developing psychological principles. By “incident,” what is meant
is any observable human activity that is sufficiently complete in itself to permit
inferences and predictions to be made about the person performing the act. To be
critical, an incident must occur in a situation where the purpose or intent of the act
seems fairly clear to the observer and where its consequences are sufficiently definite
240 ◾ Analytics and Knowledge Management

to leave little doubt concerning its effects. In this way, role-specific behavior can be
valued by observers, including their transactions with other stakeholders.
The critical incident technique outlines procedures for collecting observed inci-
dents that have special significance and meeting systematically defined criteria, for
example, behavior considered to be most effective or efficient in a certain situ-
ation (Butterfield, Borgen, Amundson, & Maglio, 2005). The many and varied
applications of the technique comprise counseling (Hrovat & Luke, 2015), tourism
(Callan, 1998), service design (Gremler, 2004), organizational merger (Durand,
2016), and healthcare (Eriksson, Wikström, Fridlund, Årestedt, & Broström,
2016). In contrast to repertory grids, which are grounded on introspection, the
critical incident technique relies on making observations about other people, thus
valuing the behavior of others. In doing so, people make notes when observing
others’ behavior to reconstruct the meaning later.
Storytelling has evolved from systemic representations of narratives (Livo &
Rietz, 1986), today termed mainly digital storytelling, as personal stories are kept in
digital form to preserve relevant happenings and share via the web (Lundby, 2008).
A story represents the world around us as perceived by specific individuals (story-
teller as creator and filter). Depending on its purpose, a story can be more like an
objective or fact-driven report or a dedicated trigger for change addressing persons
in certain social or functional roles (Denning, 2001). Digital storytelling, due to its
narrative and social structure, is used to capture not only individual characteristics
(Cavarero, 2014) but also workplace specifics (Swap, Leonard, Shields, & Abrams,
2001) and organizational issues (Boje, 1991). Similar to repertory grids, stories are
created from introspection, although they result in a linear text in contrast to a
matrix representing individual value systems.
The interviews with the value management experts lasted around 30 minutes each.
Since they had various backgrounds (marketing, information systems, psychology, lin-
guistics etc.) and diverse work experience, most of the analysts could identify methods
not included in the selection offered in item 5 of the interview guide. However, they
agreed that repertory grids are the most effective, since they constitute a noninvasive
method used to elicit value systems as the externalization procedure (see next section)
does not direct the way of answering the respective questions or thinking. A personal
construct can be understood as the underlying mechanism of a personality that enables
an individual to interpret a situation. In addition to individual attitudes, repertory
grids enable the expression of how individuals interpret a task procedure or technical
system feature, or perceive others’ behavior in a certain situation. Recording an indi-
vidual’s point of view concerning an object or phenomenon as well as the individual’s
contrasting viewpoint constitutes an individual’s value system. Hereby, a phenom-
enon is likely to have a certain meaning for each individual as expressing it, enabling
even varying experiences of situations to be represented. Yielding insights in such a
way about an individual’s value system may prompt a discussion regarding whether
a certain behavior or fact can be accepted by members of an organization or not. The
repertory grid technique therefore allows for the maximum bandwidth of responses.
 Transactional Value Analytics in Organizational Development ◾ 241

Moreover, when utilizing repertory grid elicitation, responses are expressed in

the stakeholder’s own words. However, it encourages others to delve further into
the responses to understand better and thus elicit richer information. This allows
for a thorough understanding of stakeholders’ perceptions. Finally, the rich data
collected enables a thorough examination of the underlying meaning behind each
proposed transaction and role characteristic, allowing for a specification of value
exchanges to be derived. In this way, a stakeholder can also identify any new char-
acteristics not evident in prior holomaps to become part of an improved organiza-
tional work practice.
Based on the findings and arguments brought up when discussing repertory
grids, it was decided to test this method in the context of VNA, to identify mis-
matches and matches between individual value systems and VNA-created values
influencing the flow between participants of a value network.

Externalizing Value Systems through Repertory Grids

The field test targeted the individual value systems involved in the already exempli-
fied development case of the Sales and Presales groups. The application of the reper-
tory grid technique comprises several steps (Fransella & Bannister, 1977; Fromm,
1995; Hemmecke & Stary, 2006; Honey, 1979; Jankowicz, 2004):

◾ Identification and specification of elements (objects of investigation), being

part of the considered setting.
◾ Elicitation of constructs, that is, setting up a personal construct system.
◾ Rating of constructs, that is, in a sense a consolidation of constructs.
◾ Saturation of constructs, that is, identifying the “completeness” of a construct
system.
◾ Analysis and synthesis of the result, that is, element-specific interpretation,
such as an ideal personality or object.
◾ Testing plausibility, in order to ensure a sound data basis.

The result is a validated value system in a given context that can be applied to reflect
on value creations, in the test case in the participants’ context of sales activities.

Identification and specification of elements: The elements in the repertory grid

technique are the objects of concern. Examples that can be given to the par-
ticipants for element types are physical entities (e.g., products), living things
(e.g., customers), events in a timeline (e.g., critical incidents in a customer
relationship), social entities (e.g., teams in a firm), behavior and activities
(e.g., techniques in performing a certain sales task), appraisals and abstrac-
tions (e.g., “positive” events in a customer relationship setting, criteria for
sales reports), and emotions (e.g., emotions during professional socialization).
242 ◾ Analytics and Knowledge Management

Furthermore, the participants need to be told that elements should have cer-
tain properties to support the elicitation of meaningful constructs. Ideally, they
should be: (1) discrete—the element choice should contain elements on the same
level of hierarchy, and should not contain sub-elements; (2) homogeneous—it
should be possible to compare the elements, for example, things and activities
should not be mixed within one grid; (3) comprehensible—the person from
whom the constructs are elicited should know and understand the elements,
otherwise the results will be meaningless; and (4) representative—the elicited
construct system will reflect the individually perceived reality, once the element
choice is representative for the domain of investigation.
The focus of the field test was work behavior and thus stakeholders acting
in particular roles, particularly those involved in the setting addressed by the
VNA holomapping. These are likely to form the repertory grid’s elements. In
the field test, the four participants were asked to identify between five and
eight elements as entities they could refer to in the business case addressed by
the holomap. These elements should allow naming certain properties of the
situations relevant for meeting the objectives of that business case. Since these
elements (e.g., roles, persons, objects, and task settings) stem from a running
business operation involving actual stakeholders, the participants were briefed
to use anonymous abbreviations, labels, or names when denoting the ele-
ments. The participants were also asked to include an element they consider to
be ideal in the addressed business case, to gain an understanding of the antici-
pated improvements of the transactional behavior through value creation.
Elicitation of constructs: When it comes to identifying the constructs (i.e., prop-
erties of the identified elements), we need to be aware that they are generated
through learning about the world when acting in the world. Hence, construct
systems, and thus value systems, depend on the experiences individuals make
during their lifetime and, moreover, depend on the common sociocultural
construct systems. Since construct systems change over time, they need to
be considered dynamic entities. As such, elicited knowledge through grids is
only viable for a certain person in a certain physical and social context.
Elicitation of constructs in a repertory grid interview in the field test was
concerned with understanding the participant’s perception or personal con-
struct system of work behavior. Constructs may be elicited by using one, two,
or three elements at a time. We used the triad form, that is, three out of
the specified five to eight elements during each elicitation iteration (see also
Table 8.3, presenting a repertory grid for Sales participant X who identified A,
B, C, D, and I (Ideal) as elements).
Our participants’ sample reflects an average age of 29 years (median of 33),
with the youngest being 24 and the oldest being 60. There were slightly more
men than women (three vs one). According to the objectives, the participants
have a tight range of employment, from Sales to Presales. All participants
have lived in Austria for a substantial part of their life.
 Transactional Value Analytics in Organizational Development ◾ 243

Table 8.3 Sample Repertory Grid of a Sales Person (Participant X)

Sales X
A B C D I
Construct Contrast

Openness 1 1 −3 −2 1 Information hiding

Openness 1 1 −2 3 −3 Formal reporting

Upfront information −1 2 1 2 3 Unprepared service

Informed about product 2 −2 1 2 3 Outdated product

education

Interested in method −1 −1 1 −1 1 Mainly interested in

development method application

In the triad elicitation, three elements are compared with each other
according to their similarities and differences. The person is asked to specify
“some important way in which two of them are alike and thereby different
from the third” (Fransella & Bannister, 1977, p. 14). The elicited similarity
between the two elements is recorded as the construct. Subsequently, the
person has to detail the contrast with the following question: “In what way
does the third element differ from the other two?” The construct (pair) elici-
tation continues as long as new constructs can be elicited. Table 8.3 shows a
sample grid that was elicited from Sales participant X following the elicita-
tion steps 1 to 3.
Step 1: Select triad: Each element selected by the interviewed participant was
written on a single card. The front of each card showed the name of the
respective element. In the case at hand, to ensure that the participants
could remember the addressed elements when using pseudonyms, abbre-
viations, or symbols, they were asked to make a coding list before starting
the construct elicitation procedure. After that, the interviewer turned the
index cards upside-down and mixed them, and the participant was asked
to pick three of the index cards, at random. This move prevented the
participant from knowingly selecting a specific card.
Step 2: Elicitation of raw constructs: After the participant had selected three
index cards, the participant was asked: “In the context of sales, how are
two of these elements the same, but different from the third?” The par-
ticipant expressed his or her construct system concerning the character-
istic that differentiates the selected elements. The participant was then
asked to provide a short label to describe the emergent pole (how two
are the same). For instance, Sales participant X identified “openness” for
the A and B elements in the first elicitation round (data lines 1 and 2,
244 ◾ Analytics and Knowledge Management

respectively, of Table 8.3), however these were contrasted by two con-

structs, namely “information hiding” and “formal reporting.”
Step 3: Laddering: As the participant provided insights into his or her per-
sonal construct system, the interviewer applied the laddering technique
by asking “how” and “why” questions to encourage the participant to
provide more information in areas where certain ambiguities still existed
(“openness”). A construct with two contrasting constructs was indicated
in the first two lines of the table. The laddering technique incited the
participants to pose contrasting constructs to the emergent pole. The fol-
lowing example and transcript illustrate the laddering technique.
The second participant provided the emergent pole for two role carriers. He
explained how performing formal reporting in contrast to openness induces
a feeling of not being open to each other. He declared that the feeling was
present in regards to the third person (see line 2 in Table 8.3). The interviewer
asked for an elaboration of the contrasting pole, and the second participant
replied as follows:

If you get a report without any context and personal purpose, you feel like
a mere fact checker without any background information on how a sales
process went so far and you are lost if you are supposed to arrange the next
step in customer relationship management. In particular, once the data
indicates you could have known more about a sales process when studying
the subsequent report from that person which indicates information you
should have received before launching the latest intervention.

Rating the constructs for each element: The third phase of a repertory grid elicita-
tion session is the rating of the elements according to the elicited constructs.
The mutual relations of elements and constructs can be explored by using a
rating procedure. The rating can range from 5 to 7 on a point scale. In the
case at hand, the rating ranged from +3 to −3 (excluding 0), with +3, +2, +1
for rating the strength of property (+3 is the strongest) for left-side constructs
(left row of Table 8.3), and −3, −2, −1 for rating the strength of property
(−3 is the strongest) for right-side constructs, also termed contrasts (right row
of Table 8.3). For instance, “openness” was rated for element D as very strong
(+3 in line 2), and also “formal reporting” for element I (−3, since it refers to
the contrast, rated from −3 to −1).
Construct saturation: The overall aim of construct saturation is to obtain as many
unique constructs as required to externalize the value system of a participant.
Since even people acting in the same role (e.g., Sales) in a certain organiza-
tional context are likely to have different backgrounds, there is no common
rule regarding when to stop eliciting constructs. Once all possible combina-
tions of elements have been selected, a second round of selections could make
sense, in case the participant wants to add more constructs to the already
 Transactional Value Analytics in Organizational Development ◾ 245

elicited ones in the grid. Typically, a repertory grid with four to six elements
has not more than 10 to 15 lines of entry. However, a dedicated check of
whether construct saturation has been reached is strongly advised. In the field
test this strategy was applied. We even asked the participants whether certain
constructs were considered unique to ensure that a new construct was elicited
when the participants gave a statement reflecting on a triad.
Analysis and synthesis of the result: The goal of the analysis of a repertory grid is
to represent the construct system in a way that the participant gets explicit
insights into his or her own view about the corresponding elements. Finally,
other individuals should come to an understanding about the participant’s
way of thinking. The most straightforward consideration of a grid is to look
at the ideal (I) element entries and interpret the pattern representing the ideal
case, object, or person. In addition, bi-plots derived from principal compo-
nent analysis and dendrograms derived from cluster analysis can be generated.
In bi-plots the relations of elements and constructs can be displayed, whereas
in dendrograms only the relations of elements or the relations of constructs
can be visualized. However, it is not necessary to use computer programs for
analysis, especially when the results of a single person are subject to analysis.
Content analysis (e.g., as proposed by Fromm, 1995) allows not only deter-
mining the meaning of constructs, but also clusters them according to a cat-
egory system.
Testing plausibility: A check of the plausibility of the named constructs should
be conducted with independent third parties to ensure the reliability and
stability of the results, as well as the accuracy of the data (Jankowicz, 2004).
Initially, we verified the clarity of constructs by asking two domain experts
not participating in the elicitation to read each construct carefully and explain
its meaning in his or her own words in the context of the addressed setting
(i.e., sales). The constructs were adjusted in cases where constructs were not
self-explaining. Statements from the laddering step in the elicitation phase
served as a valuable input to that task.

Developing Commitment for an Organizational Move

In this step the constructs were checked for operational improvement with respect
to the value creation in the VNA, that is, to what extent the proposed value cre-
ations represent new insights about the addressed role-specific behavior. In this
way, new patterns of exchange can be specified and verified, as the repertory grid
insights are pertinent for the development of novel role-specific behavior.
As shown in Table 8.2 (value creation analysis), for Sales participant X an impor-
tant value creation is to add value to product features received from the Product
Development group, in particular by providing timely information for interviewers
and the Customer Service group. When checking his or her repertory grid, such an
246 ◾ Analytics and Knowledge Management

improvement is in line (since positively rated for the ideal Sales participant) with
the positively rated constructs “openness,” “upfront information,” and “interested in
method development.” The same holds for the participant’s role in contributing to
timely information in cooperation with the Product Development group. Becoming
an active agent, and thus extending value through targeted requests and other efforts
to reach the Product Development group, is consistent with all positively rated con-
structs of the Ideal element. In this way, Sales participant X can check whether the
value creation is in line with his or her individual value system, before actively pro-
moting this change in the ongoing organizational development process.
In receiving new input from others (e.g., the Presales group), the same approach can
be applied. In our case, a Presales participant in the value creation analysis converted
the intangible deliverable “incomplete information” to a tangible one. Hence, a pos-
sible value creation for this Presales participant is to add value to customer relations
and product features received from the Product Development group by providing
indicators for incomplete information, since it may affect the work of interviewers
and the Customer Service group. Again, by checking his or her repertory grid, such
an improvement is totally acceptable to Sales participant X, since it is in line (i.e.,
positively rated for the Ideal Sales participant) with the constructs of “openness” and
“upfront information.” Sales participant X can support this change proposal based on
his or her elicited and rated constructs in the context of sales processes.
As these two examples reveal, the individually-valid repertory grid can be uti-
lized to check incoming and outgoing transactions as part of the value creation
analysis, and ascertain whether they fit to the individual value system. Based on
this information, value creation can either be promoted or opposed to be imple-
mented on the organizational level.

Conclusive Summary
Value congruence plays a crucial role in organizations once individual proposals
are discussed for improving business operation. We have looked at the case of indi-
vidual stakeholder transaction analyses, since VNA focuses on value creations while
targeting stakeholder- and business-conform exchange patterns. As the approach
considers both the exchange of tangible (or contracted) and intangible (or volun-
tarily) provided deliverables, value creations need to be aligned with individual value
systems before becoming a collective practice.
Various methods, in particular the repertory grid technique, the critical inci-
dent technique, and storytelling, exist to elicit individual value systems. Several
experts have been interviewed taking into account the listed candidates as well as
their experiential knowledge, to identify the best choice for eliciting value systems
in stakeholder-driven change management. They agreed on the consensus-building
process supported by individually generated repertory grids. These grids allow indi-
viduals in certain roles to reflect on their value system when acting in that role.
 Transactional Value Analytics in Organizational Development ◾ 247

In a field test, individual stakeholders were able to rethink organizational behavior

changes offered by other stakeholders, as they could assess them according to their
role-specific value system. Further studies will deepen this methodological insight,
since the reported field test only revealed the feasibility of the approach. There are
other points of interventions (e.g., the holomapping phase) when applying repertory
grids in the context of VNA. They are currently studied in ongoing field tests in other
service and production industries.

References
Allee, V. (1997). The knowledge evolution: Expanding organizational intelligence. Amsterdam,
the Netherlands: Butterworth-Heinemann.
Allee, V. (2003). The future of knowledge: Increasing prosperity through value networks.
Amsterdam, the Netherlands: Butterworth-Heinemann.
Allee, V. (2008). Value network analysis and value conversion of tangible and intangible
assets. Journal of Intellectual Capital, 9(1), 5–24.
Augl, M., & Stary, C. (2015). Communication- and value-based organizational develop-
ment at the university clinic for radiotherapy-radiation oncology. In S-BPM in the
wild (pp. 35–53). Berlin, Germany: Springer International Publishing.
Augl, M., & Stary, C. (2017). Adjusting capabilities rather than deeds in computer-
supported daily workforce planning. In M. S. Ackerman, S. P. Goggins, T. Herrmann,
M. Prilla, & C. Stary (Eds.), Designing healthcare that works. A sociotechnical approach
(pp. 175–188). Cambridge, MA: Academic Press/Elsevier.
Boje, D. M. (1991). The storytelling organization: A study of story performance in an office-
supply firm. Administrative Science Quarterly, 36(1), 106–126.
Boyle, T. A. (2005). Improving team performance using repertory grids. Team Performance
Management: An International Journal, 11(5/6), 179–187.
Butterfield, L. D., Borgen, W. A., Amundson, N. E., & Maglio, A. S. T. (2005). Fifty years
of the critical incident technique: 1954–2004 and beyond. Qualitative Research, 5(4),
475–497.
Callan, R. J. (1998). The critical incident technique in hospitality research: An illustration
from the UK lodge sector. Tourism Management, 19(1), 93–98.
Cavarero, A. (2014). Relating narratives: Storytelling and selfhood. New York, NY: Routledge.
Dalkir, K. (2011). Knowledge management in theory and practice. Cambridge, MA: MIT
Press.
Denning, S. (2001). The springboard: How storytelling ignites action in knowledge-era organi-
zations. New York, NY: Routledge.
Durand, M. (2016). Employing critical incident technique as one way to display the hid-
den aspects of post-merger integration. International Business Review, 25(1), 87–102.
Easterby-Smith, M. (1976). The repertory grid technique as a personnel tool. Management
Decision, 14(5), 239–247.
Edwards, J. R., & Cable, D. M. (2009). The value of value congruence. Journal of Applied
Psychology, 94(3), 654.
Eriksson, K., Wikström, L., Fridlund, B., Årestedt, K., & Broström, A. (2016). Patients’
experiences and actions when describing pain after surgery—A critical incident tech-
nique analysis. International Journal of Nursing Studies, 56, 27–36.
248 ◾ Analytics and Knowledge Management

Ferretti, V. (2016). From stakeholders analysis to cognitive mapping and Multi-Attribute

Value Theory: An integrated approach for policy support. European Journal of
Operational Research, 253(2), 524–541.
Firestone, J. M., & McElroy, M. W. (2003). Key issues in the new knowledge management.
New York, NY: Routledge.
Flanagan, J. C. (1954). The critical incident technique. Psychological Bulletin, 51(4), 327.
Ford, K., Perty, F., Adams-Webber, J., & Chang, P. (1991). An approach to knowledge
acquisition based on the structure of personal construct systems. IEEE Transactions
Knowledge and Data Engineering, 4(1), 78–88.
Fransella, F., & Bannister, D. (1977). A manual for repertory grid technique. London, UK:
Academic Press.
Fromm, M. (1995). Repertory grid technique: A Tutorial (in German). Weinheim, Germany:
Deutscher Studienverlag.
Gaines, B. R., & Shaw, M. L. G. (1992). Knowledge acquisition tools based on personal
construct psychology. Knowledge Engineering Review—Special Issue on Automated
Knowledge Acquisition Tools, 8, 49–85.
Goffin, K. (2002). Repertory grid technique. In D. Partington (Ed.), Essential skills for
management research (pp. 199–225). London, UK: Sage.
Goffin, K., Lemke, F., & Szwejczewski, M. (2006). An exploratory study of “close”
supplier-manufacturer relationships. Journal of Operations Management, 25, 189–209.
Goldman, A. W., & Kane, M. (2014). Concept mapping and network analysis: An analytic
approach to measure ties among constructs. Evaluation and Program Planning, 47, 9–17.
Gremler, D. D. (2004). The critical incident technique in service research. Journal of Service
Research, 7(1), 65–89.
Hemmecke, J., & Stary, C. (2006). The tacit dimension of user tasks: Elicitation and
contextual representation. In Proceedings of the International Workshop on Task
Models and Diagrams for User Interface Design (pp. 308–323). Berlin, Germany:
Springer.
Hewlin, P. F., Dumas, T. L., & Burnett, M. F. (2017). To thine own self be true? Facades
of conformity, values incongruence, and the moderating impact of leader integrity.
Academy of Management Journal, 60(1), 178–199.
Hodgkinson, G. P., Wright, R. P., & Paroutis, S. (2016). Putting words to numbers in the
discernment of meaning: Applications of repertory grid in strategic management.
In G. B. Dagnino & M. C. Cinici (Eds.), Research methods for strategic management
(pp. 201–226). New York, NY: Routledge.
Honey, P. (1979). The repertory grid in action: How to use it to conduct an attitude survey.
Industrial and Commercial Training, 11(11), 452–459.
Hrovat, A., & Luke, M. (2015). Is the personal theoretical? A critical incident analysis of
student theory journals. Journal of Counselor Preparation and Supervision, 8(1), 5.
Jankowicz, D. (2001). Why does subjectivity make us nervous? Making the tacit explicit.
Journal of Intellectual Capital, 2(1), 61–73.
Jankowicz, D. (2004). The easy guide to repertory grids. Chichester: Wiley.
Kawaf, F., & Tagg, S. (2017). The construction of online shopping experience: A repertory
grid approach. Computers in Human Behavior, 72, 222–232.
Kelly, G. A. (1955). The psychology of personal constructs: Volume one—A theory of personality.
New York, NY: Norton.
Lau, L., Yang-Turner, F., & Karacapilidis, N. (2014). Requirements for big data analytics
supporting decision making: A sensemaking perspective. In N. Karacapilidis (Ed.),
 Transactional Value Analytics in Organizational Development ◾ 249

Mastering data-intensive collaboration and decision making (pp. 49–70). Berlin,

Germany: Springer International Publishing.
Livo, N. J., & Rietz, S. A. (1986). Storytelling: Process and practice. Littleton, CO: Libraries
Unlimited.
Lundby, K. (2008). Digital storytelling, mediatized stories: Self-representations in new media
(Vol. 52). Frankfurt/Main, Germany: Peter Lang.
Novak, J. D., & Cañas, A. J. (2006). The origins of the concept mapping tool and the con-
tinuing evolution of the tool. Information Visualization, 5(3), 175–184.
Payne, A. F., Storbacka, K., & Frow, P. (2008). Managing the co-creation of value. Journal
of the Academy of Marketing Science, 36(1), 83–96.
Posner, B. Z., & Schmidt, W. H. (1993). Values congruence and differences between the
interplay of personal and organizational value systems. Journal of Business Ethics,
12(5), 341–347.
Rich, B. L., Lepine, J. A., & Crawford, E. R. (2010). Job engagement: Antecedents and
effects on job performance. Academy of Management Journal, 53(3), 617–635.
Selby, R. W. (2009). Analytics-driven dashboards enable leading indicators for require-
ments and designs of large-scale systems. IEEE Software, 26(1), 41–49.
Senior, B. (1996). Team performance: Using repertory grid technique to gain a view from
the inside. Journal of Managerial Psychology, 11(3), 26–32.
Senior, B., & Swailes, S. (2004). The dimensions of management team performance: A rep-
ertory grid study. International Journal of Productivity and Performance Management,
53(4), 317–333.
Song, S. R., & Gale, A. (2008). Investigating project managers’ work values by repertory
grids interviews. Journal of Management Development, 27(6), 541–553.
Stary, C. (2007). Intelligibility catchers for self-managed knowledge transfer. In Proceedings
of the 7th international conference on advanced learning technologies, ICALT 2007,
IEEE (pp. 517–521).
Stary, C. (2014). Non-disruptive knowledge and business processing in knowledge
life cycles—Aligning value network analysis to process management. Journal of
Knowledge Management, 18(4), 651–686.
Stary, C., Maroscher, M., & Stary, E. (2012). Knowledge management in praxis:
Methods—tools—Examples (in German). Munich, Germany: Hanser.
Stewart, V., Stewart, A., & Fonda, N. (1981). Business applications of repertory grid.
Maidenhead: McGraw-Hill.
Stoyanov, S., Jablokow, K., Rosas, S. R., Wopereis, I. G., & Kirschner, P. A. (2017). Concept
mapping—An effective method for identifying diversity and congruity in cognitive
style. Evaluation and Program Planning, 60, 238–244.
Strube, M. J. (2012). From “out there” to “in here”: Implications of self-evaluation motives
for self-knowledge. In S. Vazire & T. D. Wilson (Eds.), Handbook of self-knowledge
(pp. 397–412). New York, NY: Guilford Press.
Swap, W., Leonard, D., Shields, M., & Abrams, L. (2001). Using mentoring and story-
telling to transfer knowledge in the workplace. Journal of Management Information
Systems, 18(1), 95–114.
Tantalo, C., & Priem, R. L. (2016). Value creation through stakeholder synergy. Strategic
Management Journal, 37(2), 314–329.
Treem, J. W., & Leonardi, P. M. (2017). Recognizing expertise: Factors promoting congru-
ity between individuals’ perceptions of their own expertise and the perceptions of
their coworkers. Communication Research, 44(2), 198–224.
250 ◾ Analytics and Knowledge Management

Trochim, W. M., & McLinden, D. (2017). Introduction to a special issue on concept map-
ping. Evaluation and Program Planning, 60, 166–175.
Vargo, S. L., Maglio, P. P., & Akaka, M. A. (2008). On value and value co-creation:
A service systems and service logic perspective. European Management Journal, 26,
145–152.
Vogel, R. M., Rodell, J. B., & Lynch, J. W. (2016). Engaged and productive misfits: How
job crafting and leisure activity mitigate the negative effects of value incongruence.
Academy of Management Journal, 59(5), 1561–1584.
Wang, M., Cheng, B., Chen, J., Mercer, N., & Kirschner, P. A. (2017). The use of web-
based collaborative concept mapping to support group learning and interaction in an
online environment. The Internet and Higher Education, 34, 28–40.
Chapter 9

Data Visualization
Practices and Principles
Jeonghyun Kim and Eric R. Schuler

Contents
Introduction ......................................................................................................251
Data Visualization Practice ................................................................................253
Multidimensional Visualization ....................................................................253
Hierarchical Data Visualization ................................................................... 260
Data Visualization Principles .............................................................................265
General Principles .........................................................................................267
Specific Principles: Text.................................................................................271
Specific Principles: Color ..............................................................................271
Specific Principles: Layout ............................................................................272
Implications for Future Directions ....................................................................273
Note ..................................................................................................................274
References .........................................................................................................274

Introduction
The ever-increasing growth of Big Data has impacted every aspect of our modern
society, including business, marketing, government agencies, health care, academic
institutions, and research in almost every discipline. Due to its inherent properties
of excessive volume, variety, and velocity, it is very difficult to store, process, and
extract the intended information from it; that is, it becomes harder to grab a key
message from this universe of data. In this direction, the demand for data analytics,

251
252 ◾ Analytics and Knowledge Management

which is the process of collecting, examining, organizing, and analyzing large data
sets for scientific understanding and direct decision-making, has increased in recent
years. Various forms of analytics, including predictive analytics, data mining, and
statistical analysis, have been applied and implemented in practice; this list can be
further extended to cover data visualization, artificial intelligence, natural language
processing, and database management to support analytics.
Among those capabilities that support analytics, data visualization1 has a critical role
in the advancement of modern data analytics (Bendoly, 2016). It has become an active
and vital area of research and development that aims to facilitate reasoning effectively
about information, allowing us to formulate and test hypotheses, to find patterns and
meaning in data, and to easily explore the contours of a data collection from different
perspectives and at varying scales. As Keim et al. (2008) asserted, the goal of visualiza-
tion is to make the way of processing data and information transparent for an analytic
discourse. The visualization of these processes provides the means for communicating
about them, instead of being left with the results. Visualization fosters the construc-
tive evaluation, correction, and rapid improvement of our processes and models, and
ultimately, the improvement of the management of knowledge on all levels, including
personal, interpersonal, team, organizational, inter-organizational, and societal, as well
as enhancement of our decision-making process and management capabilities.
To achieve this goal, various visualization techniques have been developed and
proposed, evolving dramatically from simple bar charts to advanced and interactive
3D environments, such as virtual reality visualizations, for exploring terabytes of
data. A rapidly evolving landscape in the business intelligence market led to interest-
ing innovations in areas such as behavioral and predictive analytics; now visualiza-
tion is increasingly being integrated with analysis methods for visual analytics. This
integration offers powerful and immediate data exploration and understanding. The
number of tools and software solutions that integrate visualization and analysis,
including Power BI,2 Tableau,3 Qlik,4 and Domo,5 has grown rapidly during the past
several decades while keeping pace with the growth of data and data types. Toolkits
for creating high-performance, web-based visualizations, like Plotly6 and Bokeh7,
are becoming quite mature. Further, various web applications supporting visual-
ization features have emerged. For instance, web-based notebooks like Jupyter8

1 Elsewhere, other terms are being used, such as scientific visualization, information visualiza-
tion, and visual analytics in a more restricted sense (Keim et al., 2008; Tory & Möller, 2004).
In this chapter, the term data visualization will be used in a broad sense, referring to the use of
visual representations to explore, make sense of, and communicate data.
2 https://powerbi.microsoft.com
3 https://www.tableau.com
4 https://www.qlik.com
5 https://www.domo.com
6 https://plot.ly
7 http://bokeh.pydata.org
8 http://jupyter.org
 Data Visualization Practices and Principles ◾ 253

are interactive interfaces that can be used to ingest data, explore data, visualize data,
and create small reports from the results.
In recent years, discussions on data visualization have arisen, including effective
data visualization. Important questions have been raised regarding principles for data
visualization: What constitutes a good visualization? How can we ensure that we
select the best visualizations to expose the value in the data sets? These questions are
critical and deserve empirical investigation, because a good visualization can make
the process of understanding data effective, while a bad visualization may hinder the
process or convey misleading information. Thus, the purpose of this chapter is to
address these questions by providing a review of good data visualization practice tech-
niques as articulated by the leading scholars and practitioners, and discussing leading
data visualization components to ensure data visualizations are used effectively.

Data Visualization Practice

There are numerous methods to visualize data. It is critical to utilize the correct visu-
alization based on the nature of the data and the point that the analyst is trying to
present. Data visualizations can range from simple charts that portray frequencies and
percentages to complicated visualizations that provide frequencies based on a geo-
graphical location. In this section, we provide a brief overview of two broad types of
visualizations: (1) multidimensional visualization and (2) hierarchical and landscape
visualization. Multidimensional data visualization enables users to visually compare
data dimensions (column values) with other data dimensions using a spatial coor-
dinate system (Bansal & Sood, 2011); they include bar charts, distributional graphs
(i.e., histograms, boxplots, and violin plots), line graphs, scatter plots, and pie graphs.
Hierarchical and landscape data visualization differs from normal multidimensional
visualization in that they exploit or enhance the underlying structure of data set itself
(Bansal & Sood, 2011); they include both tree visualizations and map visualizations.
To better understand how different data visualizations can be utilized, we
will be referring to the following fictitious scenario: The Knowledge Management
Corporation (KMC) is a company that helps other businesses collect and analyze
various types of data to improve customer relations and profits. Recently, KMC has
been working with More Fun Toys, a national toy company, to better understand the
current trends in the market, customer feedback, and patterns of their product line
across the country. We will be referring to this example in regard to the various data
visualization tools and what types of questions each visualization tool can answer.

Multidimensional Visualization
Column charts, sometimes termed bar charts, are utilized to compare values
from different groups (Ajibade & Adeiran, 2016). More specifically, if the groups
are categorical, a bar chart can be utilized. The bars can be vertical or horizontal,
254 ◾ Analytics and Knowledge Management

depending on the number of groups and what is being presented. Bar charts help
visualize differences in the frequencies among distinct categories. Specifically,
bar charts allow individuals to accurately read numbers off the chart and view
trends across the groups (Zoss, 2016). There are some weaknesses of bar charts
that need to be noted. For example, if the categories have long titles, they can
be difficult to read. Additionally, it is important to pay attention to the numeri-
cal values of the y-axis as any axis that does not start at zero could distort the
visualization.
Another useful visualization is the line chart. Line charts are used to visual-
ize the relationship among several items over time (Ajibade & Adediran, 2016).
Additionally, various categories could be added to the visual so that each color line
reflects a different group. This allows readers to look at changes over time and across
groups. Using line graphs can be advantageous over stacked bar charts in that the
exact values can be easily read and the exact dates do not need to necessarily match
on the x-axis (Zoss, 2016). Line charts can be difficult to read if there are numerous
groups and the lines often intersect.
For example, More Fun Toys wanted to know which of their recent toys was
the most popular across the country in the last year. Since they are only inter-
ested in a visual of what toy was most frequently bought, we can use a bar chart
(Figure 9.1). Based on Figure 9.1, we can see that Super Fast Racer was the most
popular toy across the nation, closely followed by Happy Hamster and Shuffle-
bot. Furthermore, More Fun Toys was curious how three of their new toys
sold over the last four fiscal quarters; the three new toys were Happy Hamster,

20
Units sold (in thousands)

15

10

0
Action man

Clown shoes

Super Fast Racer

Barista bar fun

Existential robot

Meta-book

Woofy

Happy Hamster

Ultimate table top

Shufflebot

Figure 9.1 Bar chart that shows how many units of each toy were sold over the
2016 fiscal year.
 Data Visualization Practices and Principles ◾ 255

80 100 120
Unit sales (in thousands)
60

Super fast racer

40

Ultimate table top

Happy hamster
20

1.0 1.5 2.0 2.5 3.0 3.5 4.0

Fiscal quarter

Figure 9.2 Line graph that presents the trends in sales for three toys over each
2016 fiscal quarter.

Ultimate Table Top, and Super Fast Racer. Figure 9.2 presents the number of
units (in thousands) that each toy sold in the last four fiscal quarters. As can be
seen in Figure 9.2, the product Happy Hamster had high initial sales that dimin-
ished over time, whereas Super Fast Racer had a slow start, but sales increased
each quarter.
Usually, data analysts are interested in not only describing the data but also
making inferences from the data, termed inferential statistics (Tabachnick &
Fidell, 2007). Prior to running any sort of inferential statistics, it is important
to understand the nature of the distributions. The most insightful methods to
view the distributions are through basic data visualizations to assess for normal-
ity, outliers, and linearity of variables. Oftentimes, this starts with a histogram or
distribution graph. A histogram is akin to a bar chart in that it has each value of
a variable on the x-axis and frequencies or number of occurrences on the y-axis
(Soukup & Davidson, 2002). A histogram can inform us of how data is spread
out, whether the data is skewed positively or negatively, and whether there are
potential extreme scores or outliers within the data that could affect measures of
central tendency (i.e., mean). An additional visual for the distribution of values
for a variable is a density plot. The density plot is a smoothed histogram that can
have a normal curve superimposed over it to compare the actual distribution with
a hypothesized normal distribution. This type of visualization can help researchers
determine if the distribution of scores is normal or non-normal, as well as visually
assess for skewness. For example, More Fun Toys was interested in the distribution
of customer satisfaction scores, with scores ranging from 0 (unsatisfied) to 100
(satisfied). The distribution of scores (Figure 9.3) appears to be mostly above a score
of 50 (e.g., neutral).
The boxplot is another visualization to graphically depict the distribution of val-
ues for a single quantitative variable. This allows researchers to view aspects of the
measure’s central tendency (i.e., mean) and the distribution of values around the
256 ◾ Analytics and Knowledge Management

30
25
20
Frequency
15
10
5
0

0 20 40 60 80 100
Customer satisfaction (scale of 0 to 100)

Figure 9.3 Histogram that shows the frequencies of scores on a customer satis-
faction survey.

mean (i.e., standard deviation or variance) (Soukup & Davidson, 2002). A boxplot
is designed so that the data from the first quartile and third quartile (50% of the
data) are visualized in a box with the median dividing the box. The remaining
upper and lower 25% of the data create “whiskers” to reflect how spread out the
data are. This can be helpful to determine if there are potential floor effects (i.e.,
if many of the cases have a score of zero on the variable), which would be noted
in the visual as there being no lower whisker. Alternatively, ceiling effects, or
cases that are clustered at the highest value, could be depicted by the lack of an
upper whisker. In our scenario, More Fun Toys was interested in looking at the
marketing budget for the various products across each of the three divisions (i.e.,
Atlantic, Midwest, and West Coast). The amounts spent to advertise each product
were collapsed across the division. As can be seen in Figure 9.4, the West Coast
200 400 600 800 1000 1200
Marketing budget for 2017
(in thousands)
0

Atlantic Midwest West Coast

Divisions

Figure 9.4 Boxplot that represents the median, quartiles, and potential outliers for
the aggregated project marketing budgets for the three divisions of More Fun Toys.
 Data Visualization Practices and Principles ◾ 257

900 1000 1100

Marketing budget for 2017
(in thousands)
800
700

Figure 9.5 Violin plot that shows the quartiles and distribution of the marketing
budget for the Midwest division. The magenta color is the distribution plot that
has been mirrored on both sides. The black rectangle represents the data within
the first and third quartile. The white circle is the median and the lines extending
from the first and third quartiles are the whiskers of the boxplot. The data was
from the simulated data for the Midwest division (Figure 9.4).

division spent far less on the various marketing programs, whereas the Midwest
division far exceeded the average budget figures.
Related to boxplots are violin plots (Hintze & Nelson, 1998), which combine
a boxplot and a doubled kernel density plot. As presented in Figure 9.5, violin
plots combine the features of the density plot with the information of a boxplot
in one easy-to-interpret visualization (Hintze & Nelson, 1998). The information
about the quartiles is denoted by lines and a thin box within the mirrored density
plot. This provides more information than either plot alone, however, it is not
available within some of the commercial software packages, with the exceptions
of R and SAS.
A scatterplot can be used when there is an interest in determining if there is
a linear or non-linear relationship between two or more quantitative variables
(Tabachnick & Fidell, 2007). Simply put, the scores on one quantitative variable
(X) are plotted on the x-axis with the corresponding second variable scores (Y)
on the y-axis. If there is an ellipse or a line, then there could be a linear relation-
ship between the variables, whereas a U-shaped pattern could be indicative of a
curvilinear relationship (Howell, 2013; Tabachnick & Fidell, 2007). These visuals
are often used to assess for statistical assumptions of linearity and as a first step to
determine the relationship of two variables prior to running additional analyses
(Howell, 2013). An additional benefit to a scatterplot is that it can be used to
detect multivariate outliers, where a case has normal scores on each measure sepa-
rately, but when assessed together, the combination of both scores makes it further
removed from the other cases (Khan & Khan, 2011; Tabachnick & Fidell, 2007).
258 ◾ Analytics and Knowledge Management

10
Units sold (in thousands)
8
6
4
2
0

0 2 4 6 8 10
Number of promotions

Figure 9.6 Scatterplot that visualizes the relationship between units sold and the
number of marketing promotions in the Midwest division.

If there are more than two variables of interest, a three-dimensional scatterplot can
be utilized. The three-dimensional scatterplot provides an x, y, and z-axis, and the
corresponding three scores for each case are placed in the visual space as an orb in
the space. As shown in Figure 9.6, the Midwest division of More Fun Toys was
interested in seeing if there was a relationship between the number of units sold
and the number of promotional deals that were done in the last year. Additionally,
the Midwest division was interested in whether the units sold and promotions had
a relationship with the customer satisfaction survey that was completed by some
customers after their purchase. This was done to get an idea of whether there was
a relationship with promotions and the other two variables, and if more time and
money should be utilized on promotional offers, as presented in Figure 9.7.
When there are three variables, an alternative to the scatterplot, called a bubble
chart, can be used. Bubble charts are typically used to represent the relationships
among three variables (Ajibade & Adediran, 2016). Specifically, the relationship
between the outcome (Y) and the first variable (X1) are graphed on the x and y-axis,
like a scatter plot. Then the third variable’s relationship with the other two variables
is reflected by the size of the bubble. Like a scatterplot, each bubble represents a
single observation. Figure 9.8 shows the relationship among the number of units
sold, the number of promotions from the Midwest division, and the customer feed-
back survey that was completed by individuals after making a purchase.
Although line graphs are useful to look for trends in a continuous outcome vari-
able over time and across groups, they do not visualize the relative amounts of each
group to the whole over time. An area chart can be utilized to show changes over
time and view the groups and the total values simultaneously (Soukup & Davidson,
2002). More specifically, area graphs are used to visually compare categorical values
or groups over a continuous variable in the y-axis and how each group relates to the
 Data Visualization Practices and Principles ◾ 259

10
Customer satisfaction
8
6

10
4

8 s)
6 nd
usa
tho
2

4
2 in
d(
0 sol
0

0 2 4 6 8 10 its
Un
Number of promotions

Figure 9.7 Three-dimensional scatterplot that shows the relationship among

units sold, number of marketing promotions, and customer satisfaction for the
Midwest division.
10
8
Units sold (in thousands)
6
4
2
0

0 2 4 6 8 10
Number of advertisements

Figure 9.8 Bubble chart that shows the relationship among the units sold,
amount of marketing promotions, and customer satisfaction for the Midwest
division.

whole. In Figure 9.9, More Fun Toys wanted to determine whether it would be ben-
eficial to update their online store. The company was interested in whether the new
layout would appear more welcoming to potential customers. Furthermore, they
were curious if there were gender differences between the number of pages that were
viewed and the overall time spent on the site. To determine this, KMC conducted
260 ◾ Analytics and Knowledge Management

30

Sex
20
Count

Female
Male

10

0
0 10 20 30 40 50
Time

Figure 9.9 Area chart that visualizes the amount of time spent on searching the
retail website and the number of page views by gender.

a small pilot study and monitored how long individuals spent on the More Fun
Toys website and recorded the number of pages viewed. The amount of time and
pages clicked are shown by gender in the area chart in Figure 9.9. As can be seen
in Figure 9.9, men tended to spend less time on the website compared to women;
however, the number of pages viewed appeared to be consistent across gender.
The last frequently used multidimensional visualization technique is a pie chart
that contains wedges of data that correspond to categories or groups. Typically,
the wedges of data represent percentages of the data, with all wedges adding up to
100% (Khan & Khan, 2011). There are three subtypes of pie charts: the standard
pie chart, an exploding pie chart, and a multilevel pie chart. The exploding pie
chart separates the wedges from the rest of the wedges to visualize interactions
(Khan and Khan, 2011). Multi-level pie charts are useful when the data are hier-
archical in nature; the inner circle represents the parent level (i.e., upper level of
the nested data), and the outer circle is the child level (i.e., lower level of the nested
data). Pie charts create a visual to compare parts to the whole of one variable across
multiple categories. However, it is important to note that pie charts can make it
difficult to infer the precise value of each wedge, especially if there are numerous
categories that are then turned into very small wedges (Zoss, 2016).

Hierarchical Data Visualization

Oftentimes, individuals are clustered in a region or departments are clustered by
the type of business, making it nested or hierarchical data. When the data are
 Data Visualization Practices and Principles ◾ 261

hierarchical in nature, it is important to preserve the nested relationships since

the child group is dependent on the parent group (Bickel, 2007). One method
to visualize the hierarchical structure of the data is through tree visualizations,
which include tree maps and tree diagrams. Tree diagrams are useful in display-
ing hierarchical data as the levels of the tree diagram visually depict the location
of each piece of data in a way that quickly and intuitively conveys to a user both
the location of the data within the hierarchy and the relationship of that data to
other data in the hierarchy. To represent hierarchical data, tree maps partition
the display space into a collection of rectangular bounding boxes representing
the tree structure; they use a space-filling technique to reflect the quantitative
data (Yeh, 2010). The color and size of the boxes reflect the amounts for each of
the categories in the data (Morabito, 2016). In our scenario, the executive board
of More Fun Toys was interested to see if there are any patterns among the sales
of the Super Fast Racer toy and the stores that were nested within each state
that were also nested within the three divisions of More Fun Toys. Figure 9.10
depicts the tree map to highlight sales by the nested locations within the three
divisions.

Indiana Montana Colorado Minnesota Maine South Carolina New Jersey Arizona Wyoming

Georgia Wisconsin Kansas

Illinois Maryland Pennsylvania Rhode Island California Alaska

Mississippi South
Kentucky Nebraska Massachusetts District of Nevada New Mexico
Iowa Dakota Delaware Connecticut
Columbia

North Utah
Louisiana Ohio North
Texas Missouri Carolina Idaho
Dakota New York New
Vermont
Hampshire
Washington

Michigan Tennessee Alabama Oklahoma Arkansas West Virginia Florida Hawaii

Virginia Oregon
Midwest Atlantic WestCoast

250 300 350 400 450

Sales

Figure 9.10 Tree map that shows the number of unit sales for each state within
the three regional divisions during the 2016 fiscal year. The data is hierarchi-
cal in that states are grouped within the three divisions (Atlantic, Midwest, and
West Coast divisions). The green shading depicts the amount of sales and the
size of each box is based on the number of stores that carry Super Fast Racer in
each state.
262 ◾ Analytics and Knowledge Management

Within business information, we may be interested in not only the quantity

of purchases, but also the geographical regions in which there are a high degree
of purchases and which areas have a low market. Map visualizations provide a
way to visualize the number of purchases or other quantitative variables based
on the location of the data points (Morabito, 2016). The map visualizations can
provide information on where certain markets are meeting their quotas, whereas
others may need to use a different approach to reach the targeted audience. Map
visualizations require geographic information (e.g., ZIP codes or GPS coordi-
nates), and certain programs allow additional variables to be overlaid on the
map (e.g., gender, income levels, etc.) to provide an interactive visualization.
Interactive maps allow individuals to sort, filter, and visualize patterns within the
data. More Fun Toys was interested in viewing trends in the purchases of Super
Fast Racer by state (Figure 9.11) and seeing if certain stores sold more units than
others (Figure 9.12). An additional interest was overlaying the 2017 per capita
for each state on the map visualization. Specifically, was the toy selling better in
states with higher income?

50

40
Latitude

30

20

10
−120 −100 −80
Longitude

Figure 9.11 Map visualization that shows the units of toys sold by state.
 Data Visualization Practices and Principles ◾ 263

Store
La Rogue
Online Purchase
Price Hiker
Super Price
The Toy Chest

2017 per capita income

28,100 to 33,500
33,500 to 35,300
35,300 to 37,900
37,900 to 43,600
43,600 to 62,100

Figure 9.12 Interactive map visualization that shows the relationship among
units of the toys sold by store, geographical location, and estimated 2017 per
capita income. The filters include the five stores that carry the Super Fast Racer
(i.e., La Rogue, Online Purchase, Price Hiker, Super Price, and The Toy Chest). An
additional data layer of 2017 per capita income was added to explore for patterns
in the data.

Tree graphs can aid researchers in visualizing hierarchical data by focus-

ing on the center of the hierarchy. At the center of the hierarchy is a person,
place, or thing, termed a node, and the relationship of that node with all the
other nodes in the data are graphically represented. There are two specific types
of tree graphs: radial trees and hyperbolic trees. The radial tree focuses on split-
ting the branches from the chosen nodes equally (Morabito, 2016). It allows
the connections among the central node of the hierarchy and its relationship to
the various branches (Morabito, 2016). Furthermore, the center node can be
relocated to view the relationship of the specified node with the others. On the
other hand, hyperbolic trees utilize a nonlinear procedure to show the connec-
tions among one node and the others (Morabito, 2016). Figure 9.13 represents
a radial tree graph that visualizes the hierarchical structure of More Fun Toys’
products with a parent node of product age category (i.e., toddler, young, and
teenage), toy type (i.e., educational, action, and doll), and genre of the toy (i.e.,
horror, real life, wild west, fantasy, and space). Based on Figure 9.13, we can see
that under the toddler age group, there are two types of toys, educational and
doll. Common to both educational and doll type toys under the toddler age
group are the genres of space and real life, however, wild west-themed toys for
264 ◾ Analytics and Knowledge Management

Fantasy Real life Space

Wild west Fantasy
Space Real life
Action
Space Wild west
Educational
Doll
Real life Real life

Doll Young
Fantasy Toddler Space
Educational

Wild west Wild west

Fantasy
Space Product line
Card game
Table top
Adult Horror
Horror

Teenager Fantasy
Fantasy
Table top
Card game Horror
Wild west

Action Real life

Horror
Space
Fantasy
Wild west Fantasy
Space Horror

Figure 9.13 Radial tree graph that shows the hierarchical relationships of the
product age group, toy type, and toy genre classifications.

toddlers are only educational type toys and there are only fantasy dolls for tod-
dlers. Furthermore, only the young age group had all three types of toys. Based
on these findings, More Fun Toys could look into a future line of educational
space-themed toys for toddlers, since there are currently no toys that fall within
that classification. Additionally, More Fun Toys could add additional genres to
their adult card game selections, since they only carry two genres: a fantasy card
game and a horror card game.
Parallel coordinates plots allow the visualization of multidimensional relation-
ships of several variables rather than finite points (Inselberg & Dimsdale, 1990).
More specifically, data elements are plotted across several dimensions, with each
dimension related to a single y-axis (Khan & Khan, 2011). If a line chart or scat-
terplot is utilized with multi-dimensional data, the relationships across multiple
 Data Visualization Practices and Principles ◾ 265

Income

Humor

Age

Education
Min Max

Figure 9.14 Parallel coordinates plot that shows the relationships among educa-
tional level, age, humor level, and income.

variables can be obscured (Khan & Khan, 2011). Within a parallel coordinates
plot, individual cases can be viewed across numerous dimensions to detect patterns
in a large data set (Khan & Khan, 2011). Parallel coordinates plots can be used to
identify groups of individuals based on several variables that share the same pat-
terns (Johansson, Forsell, Lind, & Cooper, 2008). However, it is important to know
that if the parallel lines overlap, it can be difficult to identify characteristics and
subsequent patterns (Ajibade & Adediran, 2016). More Fun Toys was interested
in classifying customers into distinct groups to better market their products based
on levels of education, age, humor levels, and income. Figure 9.14 presents the
relationships among these variables in a sample of 250 customers. There was a wide
spread of education levels among the customers, with most being older based on
the lines from education to age. There appeared to be three potential groups based
on humor levels denoted by the clusters of lines from age to humor. Interestingly,
from humor scores to income, it appeared that individuals who had lower levels of
humor had an associated higher income and individuals who had higher humor
scores had lower incomes.

Data Visualization Principles

As the field of data analytics and visualization has matured, the principle of
data visualization has evolved from a single universe set into a more complex,
but still loosely defined structure. In fact, researchers in many fields have
266 ◾ Analytics and Knowledge Management

considered the problem of how to construct effective data displays for a long
time. Educational researchers who work with graphical displays are mainly
concerned with the effectiveness of charts and diagrams as instructional tools.
Human factors researchers are principally concerned with psychophysical prop-
erties of displays, especially dynamic displays, rather than the interpretation of
static displays of quantitative data (Sanders & McCormick, 1987). Designers,
statisticians, and psychologists have developed rules for constructing good
graphs and tables.
Yau (2009) asserted “Data has to be presented in a way that is relate-able;
it has to be humanized. Oftentimes we get caught up in statistical charts and
graphs, which are extremely useful, but at the same time we want to engage users
so that they stay interested, continue collecting data, and keep coming back
to the site to gauge their progress in whatever they are tracking. Users should
understand that the data is about them and reflect the choices they make in their
daily lives (p. 7).”
In this context, good data visualizations should engage the audience as well as
improve both the accuracy and the depth of their understanding; both are critical
to making smarter decisions and improving productivity. Poorly created visualiza-
tions, on the other hand, can be misleading and undermine your credibility with
your audience. They make it more difficult for them to overcome the daily data
onslaught.
The most important principle of visualizing data is to address the following two
questions: What is the message you would like to convey? and Who is your audi-
ence? You should construct data visualization so it conveys the message in the most
efficient way, and the message one wants it to convey. You also need to work closely
with your intended audience to understand what actions they want to be able to
take based on the data they consume; thus, you need to identify the needs of your
audience. If the visualization doesn’t portray what the audience needs, it does not
work. Then the next step is to choose the right visualization type.
Different types of data require different visualizations. As discussed in the pre-
vious section, line charts, for example, are most useful for showing trends over time
or a potential correlation between two variables. When there are many data points
in your dataset, it may be easier to visualize the data using a scatterplot instead.
Histograms, on the other hand, show the distribution of the data, and the shape of
the histogram can change depending on the size of the bin width.
In the next section, a general vision of graphing data, which is often consid-
ered as philosophical guidelines, is presented. The following section describes some
detailed rules and best practices to ensure data visualization is efficient. It should
be noted that the set of principles and rules for designing data visualizations pre-
sented in this chapter is neither complete nor addresses all the issues relevant in
visualization design; but it is expected that this set would provide some guidelines
for visualization designers in selecting various visualization tools and primitives to
effectively display the given data.
 Data Visualization Practices and Principles ◾ 267

General Principles
A number of scholars have published lists of principles governing the definition or
construction of a “good” visualization. Tufte (1983), who is known as a pioneer
in the field of data visualization, described graphs as “instruments for reasoning
about quantitative information (p. 10)” and argued that graphs are usually the
most effective method to describe, examine, and summarize even the largest data-
sets, and if properly designed, displays of statistical information can communicate
clearly even complex information and are one of the most powerful data analytic
tools. In his classic book, “The Visual Display of Quantitative Information,” Tufte
provided principles for graphical excellence, which is defined as follows:

◾ Graphical excellence is the well-designed presentation of interesting data of

substance, of statistics, and of design.
◾ Graphical excellence consists of complex ideas communicated with clarity,
precision, and efficiency.
◾ Graphical excellence is that which gives to the viewer the greatest number of
ideas in the shortest time with the least ink in the smallest space.
◾ Graphical excellence is nearly always multivariate.
◾ Graphical excellence requires telling the truth about the data.

Building on Tufte’s standards, Wainer (1984) published his principles for displaying
data badly, which are so clearly and humorously illustrated with real-life graphs and
tables. He defined that “the aim of good data graphics is to display data accurately
and clearly” (p. 137) and further breaks this definition into three parts: showing
data, showing data accurately, and showing data clearly, as presented in Table 9.1.
His principles provide some guidance for the design of graphs in the form of a
checklist of mistakes to be avoided. It is noteworthy that his principles address the
question of how to construct a graph, but do not address the question of why things
should be done this way.
Cleveland (1985) provides a more scientific treatment of the display of science
data that summarizes much of the research on the visual perception of graphic rep-
resentations. Unlike other scholars who took a more theorist’s intuitive approach,
Cleveland has actually tested some of his rules for making “good” graphs. He
and his colleagues have conducted many psychophysical investigations of the
elementary constituents of graphical displays by abstracting and then ordering a
list of physical dimensions. The results of those empirical works have influenced
Cleveland’s (1985) listing of the principles of graph construction. He has organized
his principles under headings that give the reader clues about how he thinks each of
his principles contributes to better data displays: clear vision, clear understanding,
scales, and general strategy, as shown in Table 9.2.
Many other authors have also reported studies attempting to figure out the best
methods of displaying data (Day, 1994; Wilkinson, 1999). Such works have been
Table 9.1 Wainer’s Twelve Principles for Displaying Data Badly
Principles Description

Showing data • Show as few data as possible.

• Hide what data you do show.

Showing data accurately • Ignore the visual metaphor altogether.

• If the data is ordered and if the visual metaphor has a natural order, a bad display will surely
emerge if you shuffle the relationship.
• Only order matters.
• For example, use length as the visual metaphor when area is what is perceived.
• Graph data out of context.
• For example, Fidell with the scale or the interval displayed in order to misrepresent the facts.

Showing data clearly • Change scales in mid-axis.

• Emphasize the trivial. (Ignore the important).
• Jiggle the baseline.
268 ◾ Analytics and Knowledge Management

• Make the graph worse by using different baseline scales for variables on the same graph.
• Austria First!
• Order graphs and tables alphabetically to obscure structure in the data that would have been
obvious had the data been ordered by some aspect of the data.
• Label: (a) Illegibly, (b) Incompletely, (c) Incorrectly, and (d) Ambiguously.
• More is murkier: (a) More decimal places and (b) More dimensions.
• If it has been done well in the past, think of another way to do it.
Table 9.2 Cleveland’s Listing of the Principles of Graph Construction
Principles Description

Clear vision • Make the data stand out. Avoid superfluity.

• Use visually prominent graphical elements to show the data.
• Use a pair of scale lines for each variable. Make the data region the interior of the rectangle formed
by the scale lines. Put tick marks outside of the data region.
• Do not clutter the data region.
• Do not overdo the number of tick marks.
• Use a reference line when there is an important value that must be seen across the entire graph,
but do not let the line interfere with the data.
• Do not allow data labels in the data region to interfere with the quantitative data or to clutter the
graph.
• Avoid putting notes, keys, and markers in the data region. Put keys and markers just outside the
data region and put notes in the legend or in the text.
• Overlapping plotting symbols must be visually distinguishable.
• Superimposed data sets must be readily visually discriminated.
• Visual clarity must be preserved under reduction and reproduction.

Clear understanding • Put major conclusions into graphical form. Make legend comprehensive and informative.
• Error bars should be clearly explained.
• When logarithms of a variable are graphed, the scale label should correspond to the tick mark
labels.
Data Visualization Practices and Principles

• Proofread graphs.
• Strive for clarity.
◾

(Continued)
269
 270 ◾

Table 9.2 (Continued) Cleveland’s Listing of the Principles of Graph Construction

Principles Description

Scales • Choose the range of the tick marks to include or nearly include the range of data.
• Subject to constraints that scales have, choose the scales so that the data fills up as much of the
data region as possible.
• It is sometimes helpful to use the pair of scale lines for a variable to show two different scales.
• Choose appropriate scales when graphs are compared.
• Do not insist that zero always be included on a scale showing magnitude.
• Use a logarithm scale when it is important to understand percent change or multiplicative factors.
• Showing data on a logarithmic scale can improve resolution.
• Use a scale break only when necessary. If a break cannot be avoided, use a full-scale break. Do not
connect numerical values on two sides of a break.
Analytics and Knowledge Management

General strategy • A large amount of quantitative information can be packed into a small region.
• Graphing data should be an interactive, experimental process.
• Graph data two or more times when it is needed.
• Many useful graphs require careful, detailed study.
 Data Visualization Practices and Principles ◾ 271

described in literature reviews; for instance, Cleveland (1987) provided a bibliogra-

phy of graphical statistics; Dibble (1997) provided a number of sources contrasting
tables and graphs; and Lewandowsky and Spence (1989) focused on the perception
of graphs.

Specific Principles: Text

Simplicity is a key principle for data visualization. To support this principle, you
should remove unnecessary text in your visualization. However, nonintrusive text
elements can help increase understanding without detracting from a visual’s mean-
ing. Such text elements include a descriptive title, subtitle, and annotation. The
title should state clearly what is being graphed or represented; it has to be short
and concise. The title should be recognizable as such because it is clearly set off
from the rest of the visualization; it should not be close enough to any line to
be grouped perceptually with it. A large font size will also prevent the title from
being perceptually grouped with other labels or parts (Kosslyn, Pinker, Simcox, &
Parkin, 1983). Additionally, subtitles need to be formatted to be subordinated to
the main title, and annotations are normally located alongside or below a specific
visualization. Both subtitles and annotations can add explanatory and interpretive
power to a visualization.
A data label identifies the name of the variable or series plotted, whereas a data
value is a number indicating a specific value in the data. Data labels should be
employed to improve the readability of data values; they need to be strategically
positioned, sized, and worded so that the text complements the visualization with-
out overshadowing it. Data labels are often displayed on or next to the data compo-
nents (i.e., bars, areas, lines) to facilitate their identification and understanding. For
data values, the best practice to is put scale values on both scales, such as the x-axis
and y-axis, and to make sure that they are closer to the correct tick mark than to
anything else (Kosslyn et al., 1983).
A font needs to be selected so that text is easy to read. A good rule of thumb is
that any font that looks fancy rather than simple should be examined closely for
readability. Multiple fonts can sometimes make sense as a tool for engaging a viewer
or emphasizing a point, but too many different fonts tend to distract from a mes-
sage. Some experts suggest three as a reasonable limit (Evergreen, 2014).

Specific Principles: Color

In data visualization, color should not be used for decorative or non-informational
purposes. Color needs to be used to show your audience what questions you are
answering as it provides information that makes for improved assessment and deci-
sion making (Campbell & Maglio, 1999); thus, the best practice is to use action
colors (e.g., bright or dark colors) to guide your audience to key parts of the dis-
play and soft colors for less important or supporting data (Few, 2012). However,
272 ◾ Analytics and Knowledge Management

you should limit the number of colors shown in a visualization because use of mul-
tiple colors can be distracting. For nominal and ordinal data types, discrete color
palettes must be used. For the bar chart, for instance, research suggests the best
option would be choosing the same color for all bars except for the bar that needs
the most attention, such as those failing below the cut score of those representing
your particular program (Evergreen & Merzner, 2013). For interval and ratio data,
one can use both discrete and continuous color palettes, according to the data
structure and to the visualization aims.
Legibility is another very important criteria that needs to be considered. The
best practice is to use solid blocks of color and avoid fill patterns, which can create
disorienting visual effects. Further, background colors should generally be white or
have very reduced colors (Tufte, 1990) but graph text should be black or dark gray
(Wheildon, 2005). Color also needs to be legible for people with color-blindness;
red–green and yellow–blue combinations need to be avoided as those colors touch
one another on the color wheel.
It is also important to note that many colors depend on, and are constrained
by, the cultural traditions or technical experience of the user (Bianco, Gasparini, &
Schettini, 2014); in fact, there are many norms. Some are more universal. For
instance, most people associate green with positive or above-goal measurements,
while judicial use of red generally indicates peril or numbers that need improve-
ment. Some are political or cultural, such as red state, blue state. Some are depen-
dent on specific domain expertise; in the black, in the red (finance).

Specific Principles: Layout

It is important to lay out the visualizations in an appealing and understandable
arrangement. Improper arrangement of visualization elements can confuse and
mislead readers. To avoid distortion in the data, the visualization should not be
too long or too narrow. The aspect ratio of a data rectangle is its height divided by
width; it is the aspect ratio that decides the shape of the visualization, so it should
be chosen with care. According to Banking to 45°, a classic design guideline first
proposed and studied by Cleveland (1994), aspect ratios that have center slopes
around 45 degrees minimize errors in visual judgments of slope ratios. Kosslyn
et al. (1983) suggested including a second vertical axis on the right of the outer
frame of the visualization if the horizontal axis is more than twice as long as the
vertical axis.
The axes should be uniform and continuous. The numbers on the visualization
increase from left to right on the x-axis and from bottom to top along the y-axis.
They increase in uniform increments, and the scales on each axis should be chosen
carefully so that the entire range of values can be plotted on the visualization. For
bar charts, histograms, line or plot charts where absolute magnitudes are impor-
tant, zero should be included on the y-axis (Robbins, 2005). When the quantitative
scale corresponds to the y-axis, it can be placed on the left side, right side, or on
 Data Visualization Practices and Principles ◾ 273

both sides of the visualization. When it corresponds to the x-axis, it can be placed
on the top, bottom, or both, but it is usually sufficient to place the scale in one place
(Few, 2012). Additionally, the numbers on the axis must follow a proper interval
in the same unit and be evenly spaced; you should not skip values when you have
numerical data.
Graphical objects inside the visualization must be sized to present ratios accu-
rately as a visualization that displays data as objects with disproportionate sizes can
be misleading. This is true for the map visualization, where a proportional symbol
is used to display absolute values. It should be noted that it may be difficult to iden-
tify the unit that the symbol refers to when the size of the symbol is bigger than
the size of corresponding spatial unit. In a plot, both length and position are better
quantitatively perceived than size, line width, or color hue, meaning that the data
values that they represent and how those values compare to other values are easily
determined (Cleveland & McGill, 1984). The order of graphic objects is important,
too. It is generally acknowledged that comparing numbers ordered alphabetically,
sequentially, or by value is best; more importantly, data should be displayed in an
order that makes logical sense to the viewer. When using bar or pie charts, you
should sort data from smallest to largest values, so they are easier to compare.
Consistent text layout and designs are also recommended to avoid distortion.
It is recommended that you avoid steep diagonal or vertical text type as it can be
difficult to read; lay out text horizontally. This includes titles, subtitles, annotations,
and data labels. Line labels and axis labels can deviate from this rule. Consistency
in the use of other elements, including line widths and box sizes, is also critical
unless those elements enable us to isolate all marks belonging to the same category.

Implications for Future Directions

It is hard to deny that data visualization is already ubiquitous. As the data rate has
increased, visualization has become more common for us to be able to get informa-
tion and insight out of the data. Advanced data visualization tools have already
offered new ways to view data and this resulting increase in data visualization has
improved our visualization literacy.
Now data visualization is entering into a new era and realm. Conceptual and
theoretical developments, new and emerging sources of intelligence, and advances
in multidimensional visualization are reshaping the potential value that analytics
and insights can provide, with visualization playing a key role. In this context, the
principles of effective data visualization will never change. But new technologies
and evolving cognitive frameworks will be opening new horizons, moving data
visualization from art to science (Towler, 2015).
However, despite the huge impact data visualization has had, it still faces
considerable challenges in the future. Here we outline important challenges in this
context.
274 ◾ Analytics and Knowledge Management

Data: Big Data has introduced an enormous opportunity to utilize visualization

tools to generate insights. Visualization applications need to accommodate
and graphically represent high-resolution input data as well as continu-
ous input data streams of high bandwidth. Additionally, data visualization
should process data from multiple and heterogeneous sources and different
data types. Further, it should be noted that more and more unstructured
data are generated but current visualization practices are more appropriate
for analyzing structured data. Thus, methods for modeling and visualizing
unstructured data need to be developed.
Infrastructure: Managing large amounts of data for visualization requires spe-
cial data structures, both in memory and on disks. Further, to fully support
the analytical process, the history of the analysis should also be recorded
and interactively edited and annotated. These requirements call for a novel
software infrastructure, built upon well understood technologies such as
databases, software components, and visualization but augmented with asyn-
chronous processing, history managements, and annotations.
Evaluation: Although this chapter provides guidelines for effective data visual-
ization, there are still clearly defined metrics or evaluation methods to mea-
sure effectiveness for tools and systems. A theoretically founded evaluation
framework needs to be developed to assess the contribution of any visualiza-
tion tools and systems toward the level of effectiveness and efficiency achieved
regarding their requirements.

Note
All the visualizations presented in this chapter used fictitious data. They were simu-
lated and created using R and Tableau.

References
Ajibade, S. S., & Adediran, A. (2016). An overview of big data visualization techniques in
data mining. International Journal of Computer Science and Information Technology
Research, 4(3), 105–113.
Bansal, K. L., & Sood, S. (2011). Data visualization: A tool of data mining. International
Journal of Computer Science and Technology, 2(3). Retrieved from http://www.ijcst.
com/vol23/1/kishorilal.pdf
Bendoly, E. (2016). Fit, bias, and enacted sense making in data visualization: Frameworks
for continuous development in operations and supply chain management analytics.
Journal of Business Logistics, 37(1), 6–17. doi:10.1111/jbl.12113.
Bianco, S., Gasparini, F., & Schettini, R. (2014). Color coding for data visualization. In
M. Khosrow-Pour (Ed.), Encyclopedia of information science and technology. Hershey,
PA: IGI Global.
 Data Visualization Practices and Principles ◾ 275

Bickel, R. (2007). Multilevel analysis for applied research: It’s just regression. New York, NY:
Guilford Press.
Campbell, C. S., & Maglio, P. P. (1999). Facilitating navigation in information spaces:
Road-signs on the World Wide Web. International Journal of Human-Computer
Studies, 50(4), 309–327.
Cleveland, W. S. (1985). The elements of graphing data. Monterey, CA: Wadsworth Advanced
Books and Software.
Cleveland, W. S. (1987). Research in statistical graphics. Journal of American Statistical
Association, 82(398), 419–423.
Cleveland, W. S. (1994). The elements of graphing data (2nd ed.). Summit, NJ: Hobart Press.
Cleveland, W. S., & McGill, M. E. (1984). Graphical perception: Theory, experimenta-
tion, and application to the development of graphical methods. Journal of American
Statistical Association, 79(387), 531–554.
Day, R. A. (1994). How to write and publish a scientific paper. Phoenix, AZ: Oryx Press.
Dibble, E. (1997). The interpretation of tables and graphs. Seattle, WA: University of
Washington.
Evergreen, S. (2014). Presenting data effectively: Communicating your findings for maximum
impact. Thousand Oaks, CA: Sage Publications.
Evergreen, S., & Merzner, C. (2013). Design principles for data visualization in evaluation.
In T. Azzam & S. Evergreen (Eds.), Data visualization, Part 2. New directions for
evaluation (pp. 5–20). New York, NY: John Wiley & Sons.
Few, S. (2012). Show me the numbers (2nd ed.). Oakland, CA: Analytics Press.
Hintze, J. L., & Nelson, R. D. (1998). Violin plots: A box plot-density trace synergism. The
American Statistician, 52(2), 181–184. doi:10.1080/00031305.1998.10480559.
Howell, D. C. (2013). Statistical methods for psychology (8th ed.). Belmont, CA: Wadsworth
Cengage Learning.
Inselberg, A., & Dimsdale, B. (1990). Parallel coordinates: A tool for visualizing
multi-dimensional geometry. In Proceedings of the 1st conference on visualization’90
(pp. 361–378). San Francisco, CA: IEEE Computer Society.
Johansson, J., Forsell, C., Lind, M., & Cooper, M. (2008). Perceiving patterns in parallel
coordinates: Determining thresholds for identification of relationship. Information
Visualization, 7(2), 152–162.
Keim, D., Andrienko, G., Fekete, J-D., Görg, C., Kohlhammer, J., & Melançon, G. (2008).
Visual analytics: Definition, process, and challenges. In A. Kerren, J. Stasko, J-D. Fekete, &
C. North (Eds.), Information visualization (pp. 154–175). Berlin, Germany: Springer.
Khan, M., & Khan, S. S. (2011). Data and information visualization methods and interac-
tive mechanisms: A survey. International Journal of Computer Applications, 34(1), 1–14.
Kosslyn, S., Pinker, S., Smicox, W., & Parkin, L. (1983). Understanding charts and graphs:
A project in applied cognitive science (NIE-40079-0066). Washington, DC: National
Institute of Education.
Lewandowsky, S., & Spence, I. (1989). The perception of statistical graphs. Sociological
Methods and Research, 18, 200–242.
Morabito, V. (2016). The future of digital business innovation: Trends and practices.
Switzerland: Springer International Publishing. Heidelberg, Germany: Springer
Verlag.
Robbins, N. (2005). Creating more effective graphs. Hoboken, NJ: Wiley-Interscience.
Sanders, M. S., & McCormick, E. J. (1987). Human factors in engineering and design
(6th ed.). New York, NY: McGraw-Hill.
276 ◾ Analytics and Knowledge Management

Soukup, T., & Davidson, I. (2002). Visual data mining: Techniques and tools for data visual-
ization and mining. New York, NY: John Wiley & Sons.
Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (6th ed.). New York,
NY: Pearson Education.
Tory, M., & Möller, T. (2004). Rethinking visualization: A high-level taxonomy.
In INFOVIS ‘04 proceedings of the IEEE symposium on information visualization
(pp. 151–158). Washington, DC: IEEE Computer Society.
Towler, W. (2015, January/February). Data visualization: The future of data visualization.
Analytics Magazine, pp. 44–51.
Tufte, E. (1983). The visual display of quantitative information (1st ed.). Cheshire, CT:
Graphic Press.
Tufte, E. (1990). Envisioning information. Cheshire, CT: Graphics Press.
Wainer, H. (1984). How to display data badly. The American Statistician, 38(2), 137–147.
Wheildon, C. (2005). Type and layout: Are you communicating or just making pretty shapes?
Mentone, Australia: The Worsley Press.
Wilkinson, L. (1999). The grammar or graphics. New York, NY: Springer-Verlag.
Yau, N. (2009). Seeing your life in data. In T. Segaran & J. Hammerbacher (Eds.), Beautiful
data. Farnham, CA: O’Reilly.
Yeh, R. K. (2010). Visualization techniques for data mining in business context: A compara-
tive analysis. Retrieved from http://www.swdsi.org/swdsi06/proceedings06/papers/
kms04.pdf
Zoss, A. M. (2016). Designing public visualizations of library data. In L. Magnuson (Ed.),
Data visualization: A guide to visual storytelling for libraries (pp. 19–44). Lanham,
MD: Rowan & Littlefield.
Chapter 10

Analytics Using
Machine Learning-
Guided Simulations
with Application to
Healthcare Scenarios
Mahmoud Elbattah and Owen Molloy

Contents
Introduction ......................................................................................................279
Motivation ....................................................................................................... 280
Simulation Modeling and Machine Learning: Toward More Integration ...... 280
The Prospective Role of Machine Learning in Simulation Modeling .............281
Related Work ....................................................................................................282
Hybrid Simulations ......................................................................................282
Artificial Intelligence-Assisted Simulations ....................................................283
Simulation-Based Healthcare Planning .........................................................283
Background: Big Data, Analytics, and Simulation Modeling ............................ 284
Definitions of Big Data................................................................................ 284
Characteristics of Big Data........................................................................... 284
Analytics .......................................................................................................287
Simulation Modeling ....................................................................................288
What Can Big Data Add to Simulation Modeling? .......................................290

277
278 ◾ Analytics and Knowledge Management

Analytics Use Case: Elderly Discharge Planning ................................................291

Case Description ..........................................................................................291
Questions of Interest.....................................................................................292
Overview of Analytics Approach .......................................................................293
Data Description ...............................................................................................293
Unsupervised Machine Learning: Discovering Patient Clusters .........................296
Outliers Removal ..........................................................................................296
Feature Scaling..............................................................................................297
Features Extraction .......................................................................................297
Clustering Approach .....................................................................................298
Selected Features ...........................................................................................298
Clustering Experiments ................................................................................298
Exploring Clusters ....................................................................................... 300
Modeling Cluster-Based Flows of Patients .........................................................302
Initial System Dynamics Model ....................................................................302
Model Assumptions and Simplifications .......................................................302
Cluster-Based System Dynamics Model ........................................................303
Modeling Patient’s Care Journey........................................................................305
Simulation Approach ....................................................................................305
Generation of Patients ..................................................................................307
Model Implementation .................................................................................307
Supervised Machine Learning: Predicting Care Outcomes................................ 308
Overview of Predictive Models .................................................................... 308
Significance of Predictive Models ..................................................................309
Training Data ...............................................................................................309
Data Preprocessing........................................................................................309
Feature Extraction .................................................................................... 310
Tackling Data Imbalances ........................................................................ 310
Feature Selection........................................................................................... 311
Learning Algorithm: Random Forests ...........................................................312
Predictors Evaluation ....................................................................................312
Results and Discussion ...................................................................................... 314
Model Verification and Validation ..................................................................... 316
Model Verification ........................................................................................ 316
Model Validation .......................................................................................... 317
Future Directions .............................................................................................. 317
Study Limitations .............................................................................................. 318
Conclusions ...................................................................................................... 319
References ......................................................................................................... 319
 Analytics Using Machine Learning-Guided Simulations ◾ 279

Introduction
Simulation modeling (SM) was considered a standalone discipline that encom-
passed designing a model of an actual or theoretical system, executing the model on
a digital computer, and analyzing the execution output (Fishwick, 1995). Based on
a virtual environment, simulation models provide extended capabilities to model
real systems in a flexible build-and-test manner. In this respect, Newell and Simon
(1959) asserted that the real power of the simulation approach is that it provides
not only a means for stating a theory, but also a very sharp criterion for testing
whether the statement is adequate. Similarly, Forrester (1968) emphasized in one of
his principles of systems that simulation-based solutions present as the only feasible
approach to represent the interdependence and nonlinearity of complex systems,
whereas analytical solutions can be impossible.
The complexity of systems can be interpreted in terms of several dimensions.
One possible dimension can be attributed to the data and metadata that represent
the system knowledge. For instance, the data complexity can correspond to one,
or more, of the four aspects that characterize the notion of Big Data including:
volume, velocity, variety, or veracity. In such scenarios, further burdens can be
unavoidably placed on the modeling process, which go beyond human capabilities.
Recently, the community of systems modeling and simulation has started
to consider the potential opportunities and challenges facing the development
of simulation models in an age marked by data-driven learning. For instance, a
study (Taylor et al., 2013) introduced the term “big simulation” to describe one
of the grand challenges for the simulation research community. Big simulation
is intended to address issues of scale for big data input, very large sets of coupled
simulation models, and the analysis of big data output from these simulations, all
running on a highly-distributed computing platform. Another more recent posi-
tion paper (Tolk, 2015) envisioned that the next generation of simulation models
will be integrated with machine learning (ML), and deep learning in particular.
The study argued that bringing modeling, simulation, Big Data, and deep learn-
ing all together can create a synergy delivering significantly improved services to
other sciences.
In line with that direction, the chapter endeavored to spur a discussion on how
the practice of modeling and simulation can be assisted by ML techniques. The
initial discussion focused on why and how Big Data and ML can provide further
support to simulations. Subsequently, a practical scenario is presented in relation
to healthcare planning in Ireland to demonstrate the applicability of our ideas.
First, unsupervised ML was utilized in a bid to discover potential patterns. The
knowledge learned by ML was then used to build simulation models with a higher
level of confidence. Second, simulation experiments were conducted with the guid-
ance of ML models trained to make predictions on the system’s behavior. The key
idea was to realize ML-guided simulations during the phases of model design or
experimentation.
280 ◾ Analytics and Knowledge Management

This chapter can be viewed as structured into two main parts as follows. The
first part initiates a discussion regarding the prospective integration of simulation
models and ML in a broader sense. That discussion is intended to serve as an open-
ing to the rationale underlying our approach. Starting in Section 5, the second part
provides a more practical standpoint for integrating SM and ML through a realistic
use case.

Motivation
Simulation Modeling and Machine
Learning: Toward More Integration
The fields of SM and ML are long-established in the world of computing. However,
both of them tended to be employed in separate territories with limited, if any, inte-
gration. This lack of integration might be attributed to a couple of issues.
First, the development of ML models is highly data-driven compared to SM.
Simulation models have been largely developed with the aid of domain experts. Hence,
the subjective expert-driven knowledge can stipulate to a great extent the behavior of
a simulation model in terms of structure, assumptions, and parameters. On the other
hand, ML models can be developed with slight involvement, if any, of experts.
Second, SM and ML were often considered to be addressing different types
of analytical questions. From the perspective of data analytics, ML is largely con-
cerned with predicting what is likely to happen. However, SM goes beyond that
question and addresses further questions for more complex scenarios (e.g., “What
if?” or “How to?”). Figure 10.1 illustrates the position of SM and ML within the

Describe Diganose Predict Prescribe

What is happening? Why did it happen? What is likely to What should be

happen? done?
Descriptive statistics Business intelligence Machine learning Simulation/
reports/scorecards experiments design optimization

Analytical sophistication

Figure 10.1 The spectrum of data analytics. (Adapted from Barga, R. et al.,
Predictive Analytics with Microsoft Azure Machine Learning, Apress, 2015;
Maoz, M., How IT should deepen big data analysis to support customer-centricity,
Gartner G00248980, 2013.)
 Analytics Using Machine Learning-Guided Simulations ◾ 281

landscape of data analytics. The figure classifies data analytics into four categories
with an increasing level of analytical sophistication.

The Prospective Role of Machine Learning

in Simulation Modeling
“Even though the assumptions of a model may not literally be exact and
complete representation of reality, if they are realistic enough for the
purpose of our analysis, we may be able to draw conclusions which can
be shown to apply to the world.” (Cohen and Cyert, 1965)
The “realism” of simulation models has always been a controversial issue. It has
been continuously argued (Shreckengost, 1985; Stanislaw, 1986) that all models,
including simulation models, are considered invalid regarding complete reality.
However, a simulation model does not need to be a complete representation to be
useful for analytics purposes, as stated earlier. A simulation model should largely
attempt to represent a sufficient level of reality, which can be acceptable regarding
the questions of interest.
In this regard, the integration of SM with ML may help simulations attain a
higher level of model realism. Our main view is that simulation experiments can be
supported by predictive models trained to carefully make predictions, which reflect
the real world with a relatively lower degree of uncertainty. To explain further, it
is assumed that a new state of the system can generate new data, as illustrated in
Figure 10.2. Based on new data (i.e., feedback), ML models can be trained to pre-
dict the future behavior of the system. The ML models can be continuously trained
and refitted to echo feedback, and update the system knowledge. In this manner,

Decision
Action

Information about State or

state of the system condition of the system

Predict Collect
behavior

Train ML Data reflecting

models new system state

Update knowledge

Figure 10.2 Data-driven feedback loop.

282 ◾ Analytics and Knowledge Management

the new system states can be “learned” aided by ML models, which can in turn help
simulation models become dynamic and more realistic.
From a more practical standpoint, ML can be utilized to predict the behavior
of system variables that may not be feasible to express analytically. For example,
Zhong et al. (2016) trained ML models within a use case for crowd modeling and
simulation. The ML models were used to learn and predict the flow of crowds.
Likewise, unsupervised ML techniques (e.g., clustering) can be used to learn about
key structural characteristics of systems, especially if tackling big data scenarios.
In a broader sense, ML can be employed as an assistive tool to help reduce
the epistemic uncertainty (Oberkampf et al., 2002) underlying simulation models.
This kind of uncertainty can be attributed to the subjective interpretation of system
knowledge by modelers, simulationists, or subject matter experts.

Related Work
Owing to the multifaceted nature of the presented work, we believe that the study
can be viewed from different perspective. Therefore, we reviewed studies with rel-
evance to the following: hybrid simulations, artificial intelligence (AI)-assisted
simulations, and simulation-based healthcare planning.

Hybrid Simulations
This study can be viewed from the perspective of developing hybrid simulations. As
suggested by Powell and Mustafee (2014), a hybrid modeling and simulation study
refers to the application of methods and techniques from disciplines like operations
research, systems engineering, or computer science to one or more stages of a simulation
study. Likewise, the study here attempted to integrate simulation models with a method
from the computer science discipline (i.e., ML). Viewed this way, we aimed to review
examples of hybrid studies that incorporated simulation methods with ML techniques.
To focus our search, two main sources were selected for review over the past 10 years
(i.e., 2016–2007): Winter Simulation conference and ACM SIGSIM Conference on
Principles of Advanced Discrete Simulation (PADS) conference. It is acknowledged
that other relevant studies could have been published in other conferences or journals,
but we believe that the selected venues provided excellent, if not the best, sources of
representative studies in accordance with the target context.
One good example is Rabelo et al. (2014) who applied hybrid modeling, where
SM and ML were used altogether in a use case related to the Panama Canal opera-
tions. A set of simulation models was developed to make predictions about the
future expansion of the canal. This information was further used to develop ML
models (e.g., neural networks and support vector machines) to help with the analy-
sis of the simulation output. With a comparable hybrid approach, Elbattah and
Molloy (2016) embraced an approach that integrated SM with ML with application
to a healthcare case. The study claimed that the use of ML improved the predictive
 Analytics Using Machine Learning-Guided Simulations ◾ 283

power of the simulation model. Another example is Zhong et al. (2016) who uti-
lized ML to assist with crowd modeling and simulation. The ML models were used
to learn and predict the flow of crowds.

Artificial Intelligence-Assisted Simulations

In an early insightful suggestion, Shannon (1975) envisioned that:
“The progress being made in AI technology opens the door for a
rethinking of the SM process for design and decision support.”
Considering ML as one of the cornerstones of AI, the chapter should also be
viewed from that standpoint. The promise of AI-assisted simulations was that the
model behavior can be generated or adjusted aided by AI techniques. Examples of
approaches were developed in this regard including knowledge-based simulations
and qualitative simulations.
Fishwick and Modjeski (2012) defined knowledge-based simulation as the appli-
cation of knowledge-based methods within AI to the field of computer simulation.
As such, the knowledge about the model can be expressed as rule-based facts that
characterize experimental conditions. Early endeavors of knowledge-based simu-
lation started with building Prolog-based simulation languages such as TS-Prolog
(Futó and Gergely, 1986), and T-CP (Cleary et al., 1985). A more recent example
of knowledge-based simulations is Lattner et al. (2010) who developed an approach
for knowledge-based adaptation of simulation models. The approach was used as a
means to enable an automated generation of model variants, with a specific focus on
automated structural changes in simulation models.
On the other hand, qualitative simulations followed a different path, which was
introduced as an attempt to replicate the process of human reasoning (i.e., qualita-
tive variables) into computer simulations. Cellier (1991a) defined qualitative simula-
tions as evaluating the behavior of a system in qualitative terms. However, Fishwick
and Modjeski (2012) described that the aim of qualitative simulations is not to adopt
qualitative methods instead of quantitative methods, but rather to use qualitative
methods as an augmentation to qualitative methods. The Qualitative Simulation
(QSIM) algorithm (Kuipers, 1986) was an early qualitative simulation study.

Simulation-Based Healthcare Planning

From the context of healthcare decision-making, the study can also be related
to similar simulation-based studies that endorsed care planning. The literature
is actually rife with examples of such studies. For instance, Harper and Shahani
(2002) used SM for the planning and management of bed capacities within
an environment of uncertainty. The simulation model was utilized to help
understand and quantify the consequences of planning and management poli-
cies. Similarly, Rashwan et al. (2013) developed a simulation model to map the
284 ◾ Analytics and Knowledge Management

dynamic flow of elderly patients in the Irish healthcare system. The model was
claimed to be useful for inspecting the outcomes of proposed policies to over-
come the delayed discharge of elderly patients. However, the literature generally
laid little emphasis on endeavors toward incorporating simulation methods and
ML techniques.

Background: Big Data, Analytics,

and Simulation Modeling
Our initial view was that the practice of modeling and simulation can be supported
by emerging technologies driven by big data analytics. For instance, ML can play a
vital role for building more realistic simulation models. This section presents a brief
background to Big Data, analytics, simulation, and the prospective opportunities
for integration.

Definitions of Big Data

Although the term Big Data has been ubiquitously used over the past years, it
has been difficult to establish a commonly accepted definition. For instance, the
Oxford Dictionary (2017) defines Big Data as “Extremely large datasets that may
be analyzed computationally to reveal patterns, trends, and associations, espe-
cially relating to human behavior and interactions.” However, different interpre-
tations of Big Data can be found based on context, domain, and computation
standards. In this regard, many studies such as Ward and Barker (2013), Beyer
and Laney (2012), Gandomi and Haider (2015), and De Mauro et al. (2015)
aimed to survey the common interpretations of Big Data in a bid to articulate a
broader definition.
In this section, we aim to review the various definitions of Big Data. In a
timeline fashion, Figure 10.3 overviews the diversity of Big Data definitions over
a timespan of more than 15 years since 1997. The early definitions of Big Data
appeared to focus on the size of data in storage. The variety of definitions clearly
signifies the multifaceted interpretations of Big Data. However, the definitions can
be considered generally in terms of: volume, complexity, and technologies.

Characteristics of Big Data

Over the years, increasing dimensions have been continuously added to distinguish
the case of Big Data. Below we present the main viewpoints that were developed for
describing the characteristics of big data.
 Analytics Using Machine Learning-Guided Simulations ◾ 285

1997 (Mashey, 1997)

Big Data term coined.

2001 (Laney, 2001)

Big Data was described in terms of volume, velocity, and variety.

2005 (Rindler, McLowry, and Hillard, 2013)

The high degree of permutations and interactions within a dataset is what defines
big data. Big Data is not a function of the size of a dataset but its complexity.

2011 (Manyika et al., 2011)

Big Data refers to datasets whose size is beyond the ability of typical database
software tools to capture, store, manage, and analyze.

2012 (Beyer, and Laney, 2012)

Big Data is high-volume, high-velocity and/or high-variety information assets
that demand cost-effective, innovative forms of information processing that
enable enhanced insight, decision making, and process automation.

(Dijcks, 2012)
Big Data is the derivation of value from traditional relational database driven
business decision-making, augmented with new sources of unstructured data.

2013 (Intel, 2013)

Big data opportunities emerge in organizations generating a median of 300 TB
of data a week. Intel asserted that the most common data type involved in
analytics is business transactions stored in relational databases followed by
documents, email, sensor data, blogs, and social media.

(Microsoft, 2013)
Big Data is the term increasingly used to describe the process of applying serious
computing power (the latest in ML and AI) to seriously massive and highly
complex sets of information.

2014 (UK Government, 2014)

Big Data refers to both large volumes of data with high level of complexity and
the analytical methods applied to them which require more advanced techniques
and technologies to derive meaningful information and insights in real time.

2015 (NIST, 2015)

Big Data is a term used to describe the large amount of data in the networked,
digitized, senso-laden, information-driven world.

Figure 10.3 Definitions of Big Data over the years.

286 ◾ Analytics and Knowledge Management

Gartner’s 3Vs: In a white paper (Laney, 2001), Big Data was characterized as hav-
ing three main attributes (i.e., 3Vs). Regarded as the basic dimensions of Big
Data, the 3Vs can be explained as follows:
1. Volume: Most organizations are currently struggling with the increasing
volumes of their data. According to an estimate by Fortune magazine
(Fortune, 2012), about five exabytes of digital data was created until
2003. The same amount of data could be created in just two days by 2011.
2. Velocity: Data velocity describes the speed at which data is created, accumu-
lated, and processed. The rapidly increasing pace of the world has placed
further demands on businesses to process information in real time or near
real time. This may mean that data should be processed on the fly or in a
streaming-based fashion to make quicker decisions (Minelli et al., 2012).
3. Variety: The assortment of data variety represents a critical factor of the
data complexity. Over the past couple of decades, data have become
increasingly unstructured as the sources of data have varied beyond the
traditional operational applications. Therefore, large-scale datasets may
likely exist in different structured, semistructured, or unstructured forms,
which can escalate the difficulty of processing tasks to a greater extent.
IBM’s 4Vs: IBM added another dimension, “veracity”, to Gartner’s 3Vs. The rea-
son behind the additional dimension was justified as that IBM’s clients started
to face data-quality issues while dealing with big data problems (Zikopoulos,
2013). Hence, IBM (2017) defined the big data dimensions as: volume, veloc-
ity, variety, and veracity. Further studies (Demchenko et al., 2014) added the
“value” dimension to IBM’s 4Vs.
Microsoft’s 6Vs: For the purpose of maximizing business value, Microsoft
extended the big data dimensions into 6Vs (Wu et al., 2016). The 6Vs included
additional dimensions for variability, veracity, and visibility. In comparison
with variety, variability refers to the complexity of data (e.g., the number of
variables in a dataset), while visibility emphasizes that there is a need to have
a full picture of data to make informative decisions. Figure 10.4 below sum-
marizes the common dimensions of Big Data as explained before.

Volume Velocity Variety Veracity Value Visibility

Gartner’s 3Vs

IBM’s 4Vs

Demchenko’s 5Vs

Microsoft’s 6Vs

Figure 10.4 The articulations of big data dimensions.

 Analytics Using Machine Learning-Guided Simulations ◾ 287

Analytics
The opportunities enabled by Big Data led to a significant interest in the practice
of data analytics. Thus, data analytics has evolved into a vibrant and broad domain
that incorporates a diversity of techniques, technologies, systems, practices, meth-
odologies, and applications.
Similar to Big Data, various definitions were developed to describe analyt-
ics. Table 10.1 presents some common definitions used to describe analytics. In
the same context, Figure 10.5 portrays the interdisciplinarity involved within

Table 10.1 Common Definitions of Analytics

Definition of Analytics Reference

The extensive use of data, statistical, and quantitative Davenport and

analysis; explanatory and predictive models; and fact- Harris (2007)
based management to drive decisions and actions. The
analytics may be input for human decisions or may drive
fully automated decisions. Analytics are a subset of what
has come to be called business intelligence.

Delivering the right decision support to the right people at Laursen and
the right time. Thorlund
(2016)

The scientific process of transforming data into insight for Liberatore and
making better decisions. Luo (2011)

Technologies Information Decision-making

Electrical engineering systems Psychology
Computer science Behavioral science

Analytics
Artificial
Operational
Intelligence
research
(machine learning)

Quantitative methods
Mathematics
Statistics
Economics (Econometrics)

Figure 10.5 A taxonomy of disciplines pertaining to data analytics. (From

Mortenson, M. J. et al., Eur. J. Oper. Res., 241, 583–595, 2015.)
288 ◾ Analytics and Knowledge Management

the practice of analytics. According to Mortenson et al. (2015), these disciplines

can fit into one or more of the following categories:

◾ Technological: Refers to the various tools used including hardware, software,

and networks, which together support the efficient processing of large-scale
datasets.
◾ Quantitative methods: Refers to the applied quantitative approaches to analyz-
ing data, such as statistics, ML, econometrics, and Operations Research (OR).
◾ Decision-making: An inherently interdisciplinary area including tools, theories,
and practices used to support and understand the decision-making process
(e.g., human–computer interaction and visualization in information systems,
or problem structuring methods in OR/MS).

Simulation Modeling
“When experimentation in the real system is infeasible, simulation
becomes the main, and perhaps the only, way to discover how complex
systems work.” (Sterman, 1994)
Shannon (1975) described simulation as the process of designing a model of
a real system and conducting experiments with this model for the purpose of
understanding the behavior of the system and/or evaluating various strategies for
the operation of the system. With a virtual build-and-test environment, simulation
provides the feasibility to model real systems that exhibit adaptive, dynamic, goal-
seeking, self-preserving, and sometimes evolutionary behavior (Meadows, 2008).
In view of the analytics spectrum as presented earlier in Figure 10.1, simula-
tion models endeavor to answer more complex questions that fall under the cat-
egory of prescriptive analytics. Therefore, SM can present as a vital component for
data analytics. This section provides a brief background of the common simulation
approaches, and the main distinctions between them.
A simulation model is developed largely based on the “world view” of a modeler.
A world view reflects how a real world system is mapped to a simulation model. In
this respect, there are three primary approaches including: system dynamics (SD),
discrete event simulation (DES), and agent-based modeling.
The SD approach assumes a very high degree of abstraction, which can be
considered adequately for strategic modeling. On the other hand, discrete-event
models maintain medium and medium–low abstraction, whereas a model com-
prises a set of individual entities that have particular characteristics in common.
Agent-based models are positioned in the middle, which can vary from very
fine-grained agents to highly abstract models. Figure 10.6 portrays the three
approaches with respect to the level of abstraction. Further, Table 10.2 makes a
more detailed comparison based on Brailsford and Hilton (2001), Lane (2000),
and Sweetser (1999).
 Analytics Using Machine Learning-Guided Simulations ◾ 289

High abstraction Aggregates, feedback loops, high level influences, ...

level
(minimum details, SD
macro level,
strategic level)
AB
System
Dynamics
Medium abstraction
level
(medium details, DE
meso level,
tactical level)
Discrete event
(process-
centric)
Low abstraction modeling Agent-based
level modeling
(maximum details,
micro level,
Individual objects, exact sizes, velocities, distances, timing...
operational level)

Figure 10.6 Simulation modeling approaches. (From Borshchev, A., The Big
Book of Simulation Modeling: Multimethod Modeling with AnyLogic 6, AnyLogic
North America, Chicago, IL, 2013.)

Table 10.2 Comparison between the SD and DES Approaches

System Dynamics Models Discrete Event Models

Scope Operational/Tactical Strategic

Nature of models Continuous/Mostly Discrete/Mostly

deterministic stochastic

World view A series of stocks and Entities flowing through

flows, in which the state a network of queues
changes are continuous. and activities.
Entities are viewed as a Entities have attributes
continuous quantity, or that describe specific
aggregates rather than features of an entity
individual entities (e.g., (e.g., entity type,
products and people). dimensions, weight,
and priority).

Building blocks Stocks, inflows, outflows, Individual entities,

tangible variables, and soft attributes, queues, and
variables. events.

Model outputs A full picture of the system. Estimates of system’s

performance.
290 ◾ Analytics and Knowledge Management

What Can Big Data Add to Simulation Modeling?

As described by Cellier (1991b), a model for a system (S) and an experiment (E) is
anything to which E can be applied to answer questions about S. In other words,
every simulation model attempts to formulate a representation of system knowl-
edge to answer questions about that system. We here review the common sources
of knowledge utilized to develop simulation models, and the potential links to Big
Data as well.
System knowledge can be learned from different sources. Huang (2013) men-
tioned four categories of knowledge that can be utilized to build a simulation
model as below:

◾ Formal knowledge: This kind of knowledge can be explicitly presented and

maintained. Examples of formal system knowledge can be theories, theo-
rems, or mathematical models. The content of formal knowledge has the
advantage of being readily accessible in its meaning and form.
◾ Informal knowledge: On the other hand, informal knowledge is implicitly repre-
sented in an intangible format in the heads of people who are familiar with the
system, (e.g., domain experts or system users). Therefore, this kind of knowl-
edge needs to be formulated using elicitation and formalization methods.
◾ Empirical data: Refers to the type of data obtained from recorded observa-
tions or measurements about the system. It potentially contains information
about the behavior of a system and its subsystems.
◾ System description: Refers to data that can contain descriptive information
about the system in terms of structure and relation information. Compared
to empirical data, system description characterizes a system and its subsys-
tems themselves. This type of knowledge is often produced by people such as
domain experts or engineers involved with designing the system. Examples
are documents, floorplans of factories, and manufacturing process maps.
Figure 10.7 sketches the sources of system knowledge.

In the era of Big Data, it should be considered that system knowledge will increas-
ingly become based on empirical data accumulated or generated autonomously.
Specifically, more data will be increasingly utilized to learn about systems. Therefore,
systems that deal with big data scenarios will inevitably place further burdens on the
modeling process, which can be beyond the capabilities of humans in many situa-
tions. For instance, the knowledge of a system can be underlying huge amounts of
data, or being accumulated with high velocity. In this regard, insightful studies such
as Tolk (2015), and Tolk et al. (2015) emphasized the need for integrating SM with
Big Data. In particular, it was stressed that big data techniques and technologies
should be considered to avail of rapidly accumulating data that may be structured
or unstructured.
 Analytics Using Machine Learning-Guided Simulations ◾ 291

System knowledge

Informal knowledge Formal knowledge

Systems observation
(mental models of (theories, theorems, etc.)
(empirical data) Systems description experts, users, etc.)
(documents, designs, etc.)

Modelers

Figure 10.7 Sources of system knowledge. (From Huang, Y., Automated

Simulation Model Generation, Doctoral dissertation, TU Delft, Delft University
of Technology, 2013.)

Analytics Use Case: Elderly Discharge Planning

For the purpose of demonstrating the applicability of the key ideas presented, a
case study was developed in relation to healthcare. The following sections elaborate
the case setting, and the development of a set of simulation and ML models. The
main goal here was to provide a practical scenario where simulation models can be
designed or guided in concert with knowledge learned with the aid of ML.

Case Description
Approaching any analytics problem needs an in-depth understanding of the
domain under study. Therefore, this section serves as a necessary background prior
to building the simulation or ML models.
In Ireland, the population has been experiencing a pronounced demographic
transition. The Health Service Executive (HSE) of Ireland reported in 2014 that
the increase in the number of people over 65 is approaching 20,000 per year (HSE,
2014b). As a result, population aging is expected to have significant impacts on a
broad range of economic and social areas, and on the demand for healthcare services.
292 ◾ Analytics and Knowledge Management

Within the context of elderly care, the focus of the case was centralized
around the care scheme of hip fracture. Hip fractures were considered for having
two-fold significance. On one hand, hip fractures represent an appropriate exam-
ple of elderly care schemes. An ample number of studies (Cooper et al., 1992;
Melton, 1996) recognized that hip fractures increase exponentially with aging,
though rates may vary from one country to another. Around 3,000 patients
per year are assumed to sustain hip fractures in Ireland (Ellanti et al., 2014).
Further, that figure may unavoidably increase due to the continuously aging
population.
From an economic perspective, hip fractures can represent a major burden on
the Irish healthcare system. According to the HSE, hip fractures were identified as
one of the most serious injuries resulting in lengthy hospital admissions and high
costs (HSE, 2014a). The median length of stay (LOS) was recorded as 13 days, and
more than two-thirds of patients are discharged to long-stay residential care after
surgery (NOCA, 2014). The cost of treating a typical hip fracture was estimated
around €12,600 (HSE, 2014a), while a different study reported a higher cost of
€14,300.

Questions of Interest
In relation to the elderly care of hip fracture, two complementary categories of
questions were addressed by the use case including: population-level questions and
patient-level questions. Table 10.3 poses the questions in detail.

Table 10.3 Questions of Interest

Scope Questions

Patient-level Q1. Given a patient’s characteristics along with care–

related factors, how to predict the inpatient length of
stay in acute facilities?

Q2. Given a patient’s characteristics along with care–

related factors, how to predict the discharge
destination?

Population-level Q3. On a population basis, what is the expected demand

for long-stay care facilities such as nursing homes?

Q4. On a population basis, what is the expected

utilization of hospital resources in terms of length of
stay?
 Analytics Using Machine Learning-Guided Simulations ◾ 293

Overview of Analytics Approach

The study embraced a multiple methodology approach that aimed to integrate SM
with ML. In particular, the approach was conducted over four stages as follows:
discovering clusters of patients, modeling clustered-flows of patients, modeling
patient’s care journey, and predicting care outcomes.
Reflecting on the questions of interest, the various stages were aimed to model the
healthcare system at two interdependent levels (i.e., population level and patient level).
Initially at the population level, we utilized unsupervised ML using clustering tech-
niques in an endeavor to discover potential underlying structures. Based on patient
records, clustering experiments were carried out to discover coherent patient clusters
that have characteristics or care outcomes in common. In light of the discovered clus-
ters, the flows of elderly patients were modeled. As such, each flow of elderly popula-
tion represented a patient cluster of specific characteristics and care-related outcomes.
Subsequently at the patient level, DES modeling was used to mimic the patient’s
care journey at a finer-grained level of details. The DES model helped to deal with
patient entities, rather than aggregate populations. Each patient entity could be treated
individually in terms of characteristics (e.g., age, gender, and type of fracture), and care-
related factors (e.g., time to admission [TTA] and time to surgery [TTS]). Eventually,
ML models were trained to make predictions on care-related outcomes for the simu-
lated patients. For every simulated patient, ML models were used to predict the LOS
and the discharge destination. Figure 10.8 illustrates the four stages of our approach.
Various tools were used through the four stages. The R language was used inten-
sively within building simulations and ML models as well. The R-package deSolve
(Soetaert et al., 2010) was particularly used to build the SD model. The deSolve
package facilitated solving the ordinary differential equations within the SD model.
The DES model was built using the R language as well.
Furthermore, the Azure ML Studio was used to develop our ML models. The
R language was used within the data preprocessing procedures. The visualizations
were mostly produced using the R-package ggplot2 (Wickham, 2009). Figure 10.9
summarizes the methods and tools used by the study.

Data Description
The main source of data used by the study is the Irish hip fracture database (IHFD)
(NOCA, 2017). The IHFD repository is the national clinical audit developed to cap-
ture care standards and outcomes for hip fracture patients in Ireland. Decisions to grant
access to the IHFD data are made by the National Office of Clinical Audit (NOCA).
The IHFD contains ample information about the patient’s journey from admis-
sion to discharge. Specifically, a typical patient record included 38 data fields such
as gender, age, type of fracture, date of admission, and LOS. A thorough explana-
tion of the data fields was available via the official data dictionary (HIPE, 2015).
294 ◾ Analytics and Knowledge Management

1. Discovering clusters of patients

Unsupervised
machine learning

Population-
scope 2. Modeling clustered-flows of patients

Patient
cluster1

Patient System dynamics

cluster2 modeling
Patient
clustern

3. Modeling patient’s care journey

Discrete-event
simulation modeling

Patient-
scope 4. Predicting care outcomes
Predicted
LOS
Prediction
model Supervised
Predicted
Elderly discharge machine learning
patient destination
Prediction
model

Figure 10.8 Approach overview.

System dynamics deSolve

models
Simulation modeling
Discrete-event
simulation models

Machine learning
Exploratory analysis and
predictive analytics
Data pre processing/ ggplot2
visualization

Figure 10.9 Tools used by the study.

 Analytics Using Machine Learning-Guided Simulations ◾ 295

The dataset included records about elderly patients aged 60 and over in particular.
The data comprised about 8,000 records over three years, from January 2013 to
December 2015. It is worth mentioning that one patient may be related to more
than one record, in cases of recurrent fractures. However, we were unfortunately
unable to determine the proportion of recurrent cases as patients had no unique
identifiers, and records were completely anonymized for the purpose of privacy.
Figure 10.10 plots a histogram of the age distribution within the dataset, while
Figure 10.11 shows the percentages of male and female patients.

800

600
Frequency

400

200

0
60 70 80 90 100
Patient age

Figure 10.10 The distribution of age within the dataset.

6000

4000
Frequency

2000

0
Female Male
Patient age

Figure 10.11 The distribution of gender within the dataset.

296 ◾ Analytics and Knowledge Management

8,000,000
Pojected population

6,000,000

4,000,000

2,000,000

0
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026
Year
Total population Elderly population >= 60

Figure 10.12 The projections of population in Ireland (2017–2026).

In addition, population projections were needed to address the population-level

questions. In this regard, the study used population information prepared by the
Central Statistics Office (CSO, 2017). The data contained comprehensive infor-
mation about the population in specific geographic areas in terms of age and sex.
However, we focused only on the elderly population aged 60 years and over, in line
with the study scope. Figure 10.12 shows the population projections in Ireland from
2017 to 2026.

Unsupervised Machine Learning: Discovering

Patient Clusters
As outlined in the approach overview, ML clustering was employed for learning
about the system structure. In this regard, the study used k-means clustering to real-
ize the segmentation of patients from a data-driven viewpoint. Specifically, elderly
patients were grouped based on the similarity of patient’s age, LOS, and elapsed
TTS. The following sections explain data preprocessing procedures, and how clus-
ters were computed.

Outliers Removal
As reported by NOCA (2014), the mean and median LOS for hip fracture
patients were recorded as 19 and 12.5 days respectively. Therefore, we only con-
sidered the patients whose LOS were no longer than 60 days to avoid the odd
influence of outliers. The excluded outliers represented approximately 5% of the
overall dataset. Figure 10.13 plots a histogram of the LOS used to identify the
outliers.
 Analytics Using Machine Learning-Guided Simulations ◾ 297

2000

1500 Count
2000
1500
Count

1000 1000
500
0
500

0
0 25 50 75 100
Length of stay (LOS)

Figure 10.13 Histogram and probability density of the LOS variable. The density
is visually expressed as a gradient ranging from green (low) to red (high). The
outliers can be observed for LOS longer than 60 days.

Feature Scaling
Feature scaling is a necessary preprocessing step in ML in cases where the range
of features values varies widely. Several studies such as Visalakshi and Thangavel
(2009), and Patel and Mehta (2011) argued that large variations within the range
of feature values can affect the quality of computed clusters. Therefore, the feature
values were rescaled to a standard range.
The min-max normalization method was used, where every feature was linearly
rescaled to the [0, 1] interval. The values were transformed using the formula below
in Equation 10.1:

x − min( x )
z= (10.1)
[max( x ) − min( x )]
Features Extraction
In a report prepared by the British Orthopedic Association (2007), six quality stan-
dards for hip fracture care were emphasised. Those standards generally reflect good
practice at key stages of hip fracture care including:

◾ All patients with a hip fracture should be admitted to an acute orthopedic

ward within 4 hours of presentation.
◾ All patients with a hip fracture who are medically fit should have surgery
within 48 hours of admission, and during normal working hours.
298 ◾ Analytics and Knowledge Management

The raw data did not include fields that explicitly captured such standards. However,
they can be derived based on the date and time values of patient arrival, admission,
and surgery. In this way, two new features were added named as TTA and “TTS”.
Eventually, only TTS was included because TTA contained a significant amount of
missing values.

Clustering Approach
The study embraced the partitional clustering approach using the k-means algo-
rithm. The k-means is one of the most widely used clustering algorithms. The
k-means clustering uses a simple iterative technique to group points in a dataset into
clusters that contain similar characteristics. Initially, a number (k) is decided that
represents centroids (center of a cluster). The algorithm iteratively places data points
into clusters by minimizing the within-cluster sum of squares as the equation below
(Jain, 2010). The algorithm converges on a solution when meeting one or more of
these conditions: the cluster assignments no longer changes, or the specified number
of iterations is completed. Equation 10.2:

J (C k ) = ∑ X −µ
X i ∈C K
i K
2
(10.2)

where:
µ K is the mean of cluster C k
J (C k ) is the squared error between µ K and the points in cluster C k

Selected Features
The k-means algorithm was originally applicable to numeric features only, where
a distance metric (e.g., Euclidean distance) can be used for measuring similar-
ity between data points. Therefore, we considered the numeric features only.
Specifically, the model was trained using the following features: LOS, age, and
TTS. However, it is worth mentioning that there are some extensions of the
k-means algorithm that attempted to incorporate categorical features, such as the
k-modes algorithm (Huang, 1998).

Clustering Experiments
As usual, the unavoidable question while approaching a clustering task is how many
clusters (k) exist? In our case, the quality of clusters was experimented using k rang-
ing from 2 to 7. Table 10.4 presents the parameters used within the clustering exper-
iments. We used the Azure ML Studio to train the clustering model.
 Analytics Using Machine Learning-Guided Simulations ◾ 299

Table 10.4 Parameters Used within the k-Means Algorithm

Parameter Value

Number of clusters (k) 2–7

Centroid initialization Random

Similarity metric Euclidian distance

Number of iterations 100

500
Within clusters sum of squares

400

300

200

1 2 3 4 5 6 7
Number of clusters (k)

Figure 10.14 Plotting the sum of squared distances within clusters.

Initially, the quality of clusters was examined based on the within cluster
sum of squares (WSS), as plotted in Figure 10.14. In view of that, it turned out
that there may be three or four well-detached clusters of patients that can best
separate the dataset. Furthermore, the suggested clusters were projected into two
dimensions based on the principal component analysis (PCA) to determine the
appropriate number of clusters, as in Figure 10.15. Each subfigure in Figure 10.16
represents the output of a single clustering experiment using a different number
of clusters (k). Initially with k = 2, the output indicated a promising tendency
of clusters, where the data space is obviously separated into two big clusters.
300 ◾ Analytics and Knowledge Management
Principal component 2

Principal component 2

Principal component 2
0.5 0.5 0.5

0.0 0.0 0.0

−1.5 −1.0 −0.5 0.0 −1.5 −1.0 −0.5 0.0 −1.5 −1.0 −0.5 0.0
(a) Principal component 1 (b) Principal component 1 (c) Principal component 1
k=2 k=3 k=4
Principal component 2

Principal component 2
Principal component 2

0.5 0.5 0.5

0.0 0.0 0.0

−1.5 −1.0 −0.5 0.0 −1.5 −1.0 −0.5 0.0 −1.5 −1.0 −0.5 0.0
(d) Principal component 1 (e) Principal component 1 (f) Principal component 1
k=5 k=6 k=7

Figure 10.15 Visualization of clustering experiments with the number of clusters

ranging from 2 to 7. The clusters are projected based on the principal components.
A mixture of colors indicates less separation of clusters. The visualizations were
produced using the R-package ggplot2. (From Wickham, H., ggplot2: Elegant
Graphics for Data Analysis, Springer Science & Business Media, New York, 2009.)

Similarly for k = 3, the clusters are still well-separated. However, the quality of
the clusters started to decline when k = 4 onwards. Thus, it turned out eventu-
ally that there were three clusters that divided the dataset into coherent cohorts
of patients.

Exploring Clusters
In this section, we aim to explore the discovered clusters in a visual manner that
can reveal potential correlations or insights, which can in turn assist with simulation
model design. The clusters were particularly examined with respect to patient char-
acteristics (e.g., age), care-related factors (e.g., TTS), and outcomes (e.g., discharge
destination).
In Figure 10.16a, the inpatient LOS is plotted with respect to the three patient
clusters. At first glance, it was obvious that the patients of cluster 3 experienced
remarkably longer LOS periods compared to cluster 1 or cluster 2. In addition,
cluster 1 and cluster 2 shared a very similar distribution of the LOS variable, apart
from a few outliers in cluster 2.
 Analytics Using Machine Learning-Guided Simulations ◾ 301

60

Length of stay (days)

40

20

0
Cluster 1 Cluster 2 Cluster 3
(a) Discovered clusters of patients
LOS

10.0
100
Time to surgery (days)

7.5 90
Patient age

5.0 80

70
2.5

60
Cluster 1 Cluster 2 Cluster 3 Cluster 1 Cluster 2 Cluster 3
(b) Discovered clusters of patients (c) Discovered clusters of patients
Time to surgery (TTS) Patient age

Figure 10.16 The variation of the LOS, TTS, and age variables within the three
patient clusters.

Second, we examined the clusters with respect to the elapsed TTS. As men-
tioned in Section 8.3, the TTS has a particular significance for being one of the
quality standards for hip fracture care. Once more, the patients of cluster 3 were
observed for having a relatively longer TTS than those patients of cluster 1 and
cluster 2. Likewise, the LOS, cluster 1, and cluster 2 had a very similar distribu-
tion of the TTS. Figure 10.16b plots the TTS variable against the three clusters of
patients.
The patient age has a considerable relevance in elderly care schemes. In our
context, the possibility of sustaining hip fractures can increase significantly with
age. It turned out that cluster 2 and cluster 3 tended to have relatively older
patients rather than cluster 1. Figure 10.16c plots the age distribution within the
three clusters.
Furthermore, the clusters were inspected for possible gender-related patterns.
Figure 10.17 shows the proportions of male and female patients within clusters.
It can be clearly noticed that the number of female patients consistently exceeded
males in all clusters.
302 ◾ Analytics and Knowledge Management

2500

2000
Count of patients

1500

1000

500

0
Cluster 1 Cluster 2 Cluster 3
Discovered patient clusters
Male patients Female patients

Figure 10.17 Distribution of male and female patients within the three clusters.

Modeling Cluster-Based Flows of Patients

Initial System Dynamics Model
The initial model was intended to provide a bird’s eye view of the care scheme
of hip fractures. The model mainly aimed at capturing the relationships among
system entities in an SD manner. However, this preliminary version did not
consider the different characteristics of patients learned by the ML clustering
experiments.
Specifically, the model focused on describing the dynamic behavior pertain-
ing to the continuous growth of population aging, and the consequent implica-
tions on the incidence of hip fractures among the elderly. The model defined
the main system actors as follows: elderly patients, acute hospitals, and dis-
charge destinations (e.g., home or long-stay care facilities). The model included
a single reinforcing loop implied by patients with a fragility history, who are
susceptible to resustain hip fractures or fall-related injuries. In this regard, the
HSE (2014a) reported that one in three of elderly patients fall every year and
two-thirds of them fall again within six months. Figure 10.18 illustrates the
initial SD model.

Model Assumptions and Simplifications

A set of assumptions and simplifications was decided while maintaining the simu-
lation model as a reasonably approximate representation of the actual healthcare
system. Table 10.5 provides the assumptions and simplifications along with expla-
nations. Table 10.6 lists the model variables and Table 10.7 presents the model
equations.
 Analytics Using Machine Learning-Guided Simulations ◾ 303

Long-stay care
Hip fracture rate for discharged
elderly males
Discharge fraction
Potential male Total discharged
patients + + patients
+
New male cases
+ Inhospital
+ Home-discharged
Total elderly
population New female R
+ cases
+ Return patients
+ + +
+
Potential female
patients
Recurrence
Hip fracture rate for fraction
elderly females

Figure 10.18 Initial SD model.

Table 10.5 Model Assumptions and Simplifications

Assumption/Simplification Purpose/Reason

The rate of hip fracture in the total The rate was defined by Dodds et al.
population aged 60 and over was set (2009).
as 407 for females and 140 for males
per 100,000.

Elderly patients were assumed as To conform to the defined hip

aged 60 and over, though usually fracture rate, which endorsed
considered as aged 65 and older patients aged 60 and over.
(Rosenberg and Everitt, 2001).

The model did not consider the Only for simplification, assuming that
scenario of patient transfer from an the treatment course was bounded
acute hospital to another during within a single acute hospital.
treatment course.

The model used the same age For the purpose of simplification,
distribution for both male and since both distributions were
female elderly patients. slightly different.

Cluster-Based System Dynamics Model

The SD model was redesigned in light of the clustering experiments. In particular,
the model was disaggregated into three stocks representing the suggested clusters
of patients. Furthermore, the auxiliary variables were decided based on the cluster
analysis. For instance, the first and second patient clusters were set to undergo the
same TTS delay (i.e., TimeToSurgery1), while the third cluster was assigned a dif-
ferent delay (i.e., TimeToSurgery2).
304 ◾ Analytics and Knowledge Management

Table 10.6 Model Variables

Variable Description

Total elderly population The number of elderly population, aged

60 and over, in a particular year.

Potential male patients Total male patients aged 60 and over.

Potential female patients Total female patients aged 60 and over.

Hip fracture rate for elderly males The rate of hip fracture in the total
elderly male population = 140 cases per
100 K.

Hip fracture rate for elderly The rate of hip fracture in the total
females population aged 60 and over = 407 for
females per 100 K.

InHospital Stock variable representing the total

number of elderly hip-fracture patients
in acute hospitals nationwide.

Discharge fraction Proportion of total elderly patients

discharged to home or long-stay care.

Total discharged patients The number of patients discharged to

home and long-stay care.

Recurrence rate The rate that defines the proportion of

discharged patients who are susceptible
to resustain a hip fracture and return to
an acute hospital.

Equally important, the inflows of elderly patients were structured based on

the age variation within clusters. In particular, both of the first and third patient
clusters were modeled to include more elderly patients (i.e., ages 80–100), while
the second cluster was associated with less elderly patients (i.e., ages 60–80). This
reflected the age groups within the patient clusters. Figure 10.19 illustrates the
cluster-based SD model.
In this manner, the clustering model was employed effectively for the purpose
of understanding the system structure, where the SD model stocks actually rep-
resented the three discovered clusters of patients. Moreover, the variations within
clusters in terms of patient characteristics (e.g., age), or care-related factors (e.g.,
TTS) assisted with shaping the model behavior. As such, it can be argued that
the SD model was constructed with an established confidence predicated on the
clustering model. The well-validated quality of clusters along with the compelling
 Analytics Using Machine Learning-Guided Simulations ◾ 305

Table 10.7 Model Equations

Equation Type

1. Hip fracture rate for elderly males = 140 cases per 100,000. Auxiliary

2. Hip fracture rate for elderly females = 407 cases per 100,000. Auxiliary

3. New male cases = Hip fracture rate for elderly males * Inflow
Potential male patients

4. New female cases = Hip fracture rate for elderly females * Inflow
Potential female patients

5. Home-discharged = InHospital * Discharge fraction Outflow

6. Long-stay care discharged = InHospital * (1-Discharge fraction) Outflow

7. Recurrent patients = (Home-discharged * Recurrence rate) + Inflow

(Long-stay care discharged * Recurrence rate)

8. InHospital = Integ((New male cases + New female cases) − Stock

(Home-discharged + Long-stay care discharged) + Recurrent
patients, initial value)

visualizations could support the rationale behind the SD model design in terms of
structure and behavior as well.

Modeling Patient’s Care Journey

Simulation Approach
The DES approach was used to model a fine-grained perspective of the patient’s
journey. The model was utilized in order to produce a realistic sequence of events
corresponding to those within the care journey, as sketched in Figure 10.20.
Further, the DES approach can facilitate the following:

◾ Representing the system’s components in terms of entities (e.g., patients, hos-

pitals etc.).
◾ The entity-based modeling allowed us to consider cluster-based aspects on
the individual patient level. For instance, specific attributes of elderly patients
(e.g., fracture type) were determined differently with respect to the cluster.
◾ With such entity-based structuring, ML models can subsequently be used to
provide accurate predictions at the patient level.
 Hip fracture rate
Hip fracture rate for for elderly males
elderly females
Male
population-cluster 1 Male patients admitted 1 Cluster 1-admitted Cluster 1-in ward
Ready for surgery 1 Cluster 1-discharged
Female patients admitted 1
Female
population-cluster 1
Time to surgery 1 Time to discharge 1

Male
Elderly population-cluster 2
population Male patients admitted 2 Cluster 2-admitted Cluster 2-in ward Total
Ready for surgery 2 Cluster 2-discharged discharged
Female patients admitted 2

Female
population-cluster 2
Time to discharge 2
306 ◾ Analytics and Knowledge Management

Time to surgery 2
Male
population-cluster 3

Male patients admitted 3 Cluster 3-admitted Cluster 1-in ward1

Female Ready for surgery 3 Cluster 3-discharged
population-cluster 3 Female patients admitted 3

Figure 10.19 Cluster-based SD model.

 Analytics Using Machine Learning-Guided Simulations ◾ 307

1. Admission 2. Treatment and assessment 3. Outcome

Admitted to Admitted to
Home emergency orthopedic Home
department ward
(ED)
Primary Discharge
Nursing destination?
home surgery
Time to Time to
admission surgery
Acute Geriatric Long-stay
hospital assessment care

Non-acute Multi-
Fragility Yes
hospital disciplinary
history?
assessment

No Falls
assessment

Length of stay (LOS)

Figure 10.20 The patient journey simulated by the DES model.

Generation of Patients
The DES model made use of the projections produced by the SD model to gener-
ate individual patient entities. The generation process was implemented using the
R language. The total number of generated patients reached around 30,000 for a
simulated period of 10 years (i.e., 2017–2026). Table 10.8 presents the counts of
elderly patients generated for every cluster.

Model Implementation
The DES model was mainly developed based on the empirical data acquired by
the study. For instance, the probability distributions of patient attributes were set
to mimic reality, as in the IHFD dataset. The simulation model was fully imple-
mented using the R language. The source code can be accessed via our GitHub
repository (Elbattah, 2017).

Table 10.8 Counts of Patients Generated by the DES Model

Patient Cluster No. of Simulation-Generated Patients

Cluster 1 13,438

Cluster 2 10,782

Cluster 3 5,272
308 ◾ Analytics and Knowledge Management

Predicted
LOS DES Model
Simulated
patient
Predicted
discharge dest.

Patient
Discharge characteristics
destination
classifier

LOS
regression model

Azure predictive
web services

Figure 10.21 The experimental environment.

The main entity of the simulation model obviously represented the elderly
patient. Each patient was assigned a set of attributes that characterized age, sex, area
of residence, fracture type, fragility history and diagnosis type. The patients’ char-
acteristics varied based on the cluster they were assigned to. Further care-related
factors (e.g., TTS) were considered on an individual basis as well.
The experimental environment consisted of two integrated parts. The DES
model served as the core component. In tandem with the simulation model, ML
models were then utilized to carefully predict the inpatient LOS and discharge
destination for each elderly patient generated by the simulation model. The predic-
tions were obtained from the ML models via web services enabled by the Microsoft
Azure platform. Figure 10.21 illustrates the environment of simulation experiments
where the DES was integrated with predictions from the ML models. In this man-
ner, the ML models were employed to guide the simulation model with respect to
the LOS and discharge destination of simulation-generated elderly patients.

Supervised Machine Learning: Predicting

Care Outcomes
Overview of Predictive Models
As highlighted in Section 6, our approach endeavored to assist SM with ML. At the
outset, the unsupervised data clustering helped to design the SD model structure
 Analytics Using Machine Learning-Guided Simulations ◾ 309

and behavior to depict the real system in an accurate manner. In this section, we go
through the development of supervised ML models (e.g., regression and classifica-
tion), which were utilized to guide the DES model.
Based on the IHFD patient records, ML was employed to make predictions
on important outcomes related to the patient’s journey. At the micro level, the
ML models were aimed at addressing patient entities generated by the simulation
model. The ML models included a regression model for predicting the LOS, and
a binary classifier for predicting discharge destinations. The predicted discharge
destination included either home or a long-stay care facility (e.g., nursing home).
The ML models were developed using the Azure ML Studio.

Significance of Predictive Models

The LOS and discharge destination are of vital importance within healthcare
schemes. For instance, the LOS was considered as a significant measure of patient
outcomes, as adopted by many studies such as (O’Keefe et al., 1999), and (Englert
et al., 2001). From an operational perspective, the LOS was recognized as a valid
proxy to also measure the consumption of hospital resources (Faddy and McClean,
1999; Marshall and McClean, 2003). Other studies (Brasel et al., 2007) argued
that there was an evident correlation between the LOS and discharge destination.
Further, it was reported by Johansen et al. (2013) that the LOS represents a major
segment of the overall cost of hip fracture treatment.
Similarly, the early prediction of discharge destinations can be of significant
benefit to care planning. On a population basis, the aggregated predictions can help
estimate the desired capacity of long-stay care facilities (e.g., hospices and nursing
homes) over the simulation period (i.e., 10 years).

Training Data
As mentioned previously, the study acquired a dataset of the IHFD repository. The
IHFD dataset was used for training both the regression and classification models.
Initially, we explored the variables that can serve as features for training the ML
models. Based on our intuition, many irrelevant variables were simply excluded
(e.g., admission time and discharge time). Table 10.9 lists the variables initially
considered as candidate features.

Data Preprocessing
This section describes the data preprocessing phase conducted prior to training
the ML models. The preprocessing procedures included: removing outliers, scaling
features, tackling data imbalances, and extracting features. Removing outliers and
scaling features were already conducted before building the clustering model, as
310 ◾ Analytics and Knowledge Management

Table 10.9 Variables Explored as Candidate Features

Variables Explored

Source hospital Admission type Discharge code

Residence area Patient gender Discharge status

Admission source Hospital transferred Hospital transferred to

from

Age LOS ICD-10 diagnosis

Admission trauma type Admission via ED Fracture type

Pre-fracture mobility Fragility history Specialist falls

assessment

Multi-rehabilitation assessment

elaborated in Sections 8.1 and 8.2. Therefore, this section only explains the proce-
dures performed for tackling imbalances and extracting features.

Feature Extraction
As mentioned in Section 8.3, the TTA and TTS represent important quality care-
related factors for the hip fracture care scheme. The ML models utilized the TTA and
TTS features, which were extracted earlier during the clustering model development.

Tackling Data Imbalances

Learning with imbalanced datasets was continuously recognized as one of the prin-
cipal challenges for ML (Yang and Wu, 2006; Galar et al., 2012). The negative
impacts of imbalanced data on the prediction accuracy were emphasized by numer-
ous studies such as Japkowicz and Stephen (2002), and Sun et al. (2009).
In our case, the training data suffered from imbalanced class distributions, where a
particular class predominated the dataset. In particular, imbalanced training samples
were pronounced for patients who had LOS longer than 18 days. Likewise, imbalanced
samples could be observed for patients who were discharged to home. Moreover, training
samples for male patients, and particular age groups were obviously underrepresented.
Galar et al. (2012) identified two strategies to deal with the imbalance prob-
lem including algorithm-level approach and data-level approach, such as unders-
ampling or oversampling. The oversampling technique (Chawla et al., 2002) was
adopted in our case. The underrepresented samples were resampled at random until
they contained sufficient examples. Figures 10.22 and 10.23 show the histograms
of LOS and discharge destination respectively, where imbalanced distributions can
be clearly observed.
 Analytics Using Machine Learning-Guided Simulations ◾ 311

2500

2000
Count

1500 2000
Frequency

1500
1000
1000 500
0

500

0
0 20 40 60
LOS

Figure 10.22 Data imbalance (LOS variable).

5000

4000

3000
Frequency

2000

1000

0
Home Nursing home
Discharge destination

Figure 10.23 Data imbalance (discharge destination variable).

Feature Selection
The dataset initially contained 38 features; however, not all of them were rele-
vant. Intuitively irrelevant features were excluded. In addition, the most important
features were decided based on the technique of permutation feature importance
(Altmann et al., 2010). Table 10.10 presents the set of features used by both models.
312 ◾ Analytics and Knowledge Management

Table 10.10 Selected Features

Prediction Model Selected Features

Regression model-LOS predictor Age, patient gender, fracture type,

hospital admitted to, ICD-10
diagnosis, fragility history, TTS, TTA

Classifier model-discharge Age, patient gender, fracture type,

destination predictor hospital admitted to, ICD-10
diagnosis, LOS, fragility history, TTS

Learning Algorithm: Random Forests

The random forests algorithm (Breiman, 2001) was used for regression and classifica-
tion as well. A random forest consists of a collection of decision trees {h(x, Θk), k = 1,...},
where the {Θk} are independent identically distributed random vectors and each tree
casts a unit vote for the most popular prediction at input x. The main idea is that for
the kth tree, a random vector Θk is generated, independently of the past random vec-
tors Θ1,..., Θk−1 but with the same distribution; and a tree is grown using the training
set and Θk, resulting in a classifier h(x, Θk) where x is an input vector. Figure 10.24
portrays a simple example of a random forest composed of 3 decision trees.
Predictions are made through weighted voting for the most confident predicted class,
and the trees that have a higher prediction confidence will have a greater weight in the
final decision of the ensemble. The aggregation of voting can be done by a simple averag-
ing operation as in Equation 10.3. Our models consisted of 8 decision trees. Table 10.11
presents the parameters used for training the regression and classifier models.
T
p(c | v ) =
1
T ∑ p (c | v )
t =1
t (10.3)

where pt(c |v) denotes the posterior distribution obtained by the t-th tree.

Predictors Evaluation
The predictive models were tested using a subset from the dataset described in
Section 10.7. The randomly sampled test data represented approximately 40% of
the dataset. The prediction error of each model was estimated by applying 10-fold
cross-validation. Tables 10.12 and 10.13 present evaluation metrics of the regression
and classifier models respectively. Further, Figure 10.25 shows the Area Under the
Curve (AUC) of the classifier model.
 Analytics Using Machine Learning-Guided Simulations ◾ 313

Full
dataset

Random Random Random

subset1 subset2
… subsetK
Tree1 Tree2 TreeK

Prediction1 Prediction2 PredictionK

Majority voting

Final prediction

Figure 10.24 Random forest example: Combining predictions using majority

voting.

Table 10.11 Parameters of Random Forests

No. of decision trees 8

Max. depth of decision trees 32

No. of random splits per node 128

Table 10.12 Average Accuracy Based on 10-Fold Cross-Validation

(LOS Regression Model)
Relative absolute error ≈0.30

Relative squared error ≈0.17

Coefficient of determination ≈0.83

314 ◾ Analytics and Knowledge Management

Table 10.13 Average Accuracy Based on 10-Fold Cross-Validation

(Discharge Destination Classifier)
AUC ≈0.87

Accuracy ≈80%

Precision ≈81%

Recall ≈79%

F1 score ≈80%

1.0

0.9

0.8

0.7
True positive rate

0.6

0.5

0.4

0.3

0.2

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
False positive rate

Figure 10.25 Classifier AUC = 0.865 (discharge destination classifier).

Results and Discussion

In line with the questions of interest, the simulation model output was interpreted
in terms of the following: the expected number of discharged patients, the inpatient
LOS, and the expected demand for long-stay care facilities as a discharge destina-
tion. Figure 10.26 plots the projections of elderly patients discharged from 2017 to
2027. It can be clearly observed that cluster 1 and cluster 2 steadily included the larg-
est proportions of elderly patients. This adequately corresponded to the three clusters
discovered by the ML clustering experiments, as shown earlier in Figure 10.16.
Figure 10.28a plots a histogram for the overall LOS expected over the simu-
lated period for the three patient clusters. The histogram shows that the majority of
 Analytics Using Machine Learning-Guided Simulations ◾ 315

1800
1600
1400
Count of elderly patients

1200
1000
800
600
400
200
0
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027
Year
Patient cluster 1 Patient cluster 2 Patient cluster 3

Figure 10.26 Cluster projections.

0.06

Density
Frequency

0.06
0.04
0.04

0.02

0.00
0.02

0.00
0 10 20 30 40 50
(a) LOS
Over all LOS
0.08
0.09 0.06
0.06
Frequency
Frequency

Frequency

0.06 0.04
0.04

0.02 0.03 0.02

0.00 0.00 0.00

0 10 20 30 40 10 20 30 10 20 30 40 50
(b) LOS (c) LOS (d) LOS
LOS cluster 1 LOS cluster 2 LOS cluster 3

Figure 10.27 LOS experienced for the simulated patients.

patients experienced an LOS in the range of 10–30 days. In this regard, the model
seemed to largely mimic reality, especially after excluding outliers. Moreover,
Figures 10.27b, c, and d show the LOS with respect to every cluster individually.
It turned out that cluster 1 and cluster 3 shared a similar distribution of the inpa-
tient LOS, which to tended to be relatively longer compared to cluster 2 patients.
316 ◾ Analytics and Knowledge Management

20000

15000
Frequency

10000

5000

0
Home Nursing home
LOS

Figure 10.28 Expected demand for discharge destinations.

The similar LOS distribution might be due to including the same age group (i.e.,
aged 80–100). This can translate into the importance of considering early interven-
tion schemes for the category of more elderly patients.
Furthermore, the simulation output anticipated a significant demand for long-
stay care facilities (e.g. nursing homes) over the simulated period. As shown in
Figure 10.28, the overall number of patients discharged to long-stay care is around
22,000 compared to only 7,000 of home-discharged patients. This pronounced
difference raises an inevitable need for planning the capacity of nursing homes or
similar residential care facilities.

Model Verification and Validation

Model Verification
To validate the logic and suitability of the simulation model, a set of verification
tests (Martis, 2006) was conducted throughout the simulation model’s develop-
ment, as follows:

◾ Structure-verification test: The model structure was checked against the actual
system. Specifically, it was verified that the model structure was a reasonable
representation of reality in terms of the underlying patient clusters, and associ-
ated elderly populations.
 Analytics Using Machine Learning-Guided Simulations ◾ 317

◾ Extreme conditions test: The equations of the simulation model were tested
in extreme conditions. For example, the flows of elderly patients were set to
exceptional cases (e.g., no elderly population aged 60 or over).
◾ Parameter-verification test: The model parameters and their numerical values
were inspected if they largely corresponded to reality. Specifically, the prob-
ability distributions of patient attributes (e.g. age, sex, and fracture types)
were compared against those derived from the IHFD dataset.

Model Validation
According to Law (2008), the most definitive test of a simulation model’s validity
is comparing its outputs to the actual system. Similarly, we used the distribution of
discharge destinations as a measure of the approximation between the simulation
model and the actual healthcare system.
On one hand, Figure 10.29 provides a histogram-based comparison between
the actual system and the simulation model regarding the discharge destination.
The comparison showed that the distributions of the actual system and simula-
tion output were largely similar. However, the comparison revealed that the model
slightly underestimated and overestimated the proportion of patients discharged to
homes and long-stay care facilities respectively.

Future Directions
A further direction is to consider more sophisticated ML techniques to extract the
knowledge underlying further complex systems or scenarios. For example, it would be
interesting to investigate how simulation models can be integrated with deep learning
(DL). DL (LeCun et al., 2015) significantly helped approach hard ML problems such

0.6
0.6

0.4
Density

Density

0.4

0.2 0.2

0.0 0.0
Home Nursing home Home Nursing home
(a) Discharge destination (b) Discharge destination
Actual system Simulation model

Figure 10.29 Histograms of the discharge destination output from the actual
system and simulation model.
318 ◾ Analytics and Knowledge Management

…
Data captured at Data captured at
system state 1 system state n

Training dataset

Re-train Input variables/ Simulation Simulated

DNN to adapt parameterts model behaviour
to new system state

Features Predicted
variable

Trained deep neural network

Figure 10.30 Adapting to the new system states via predictions from a trained
DNN.

as speech recognition and visual object recognition for example. The capabilities of
DL allow computational models that are composed of multiple processing layers to
learn representations of data with multiple levels of abstraction. The multiple process-
ing layers can effectively represent linear and nonlinear transformations.
We conceive that simulation models can be assisted by predictions from deep
neural networks (DNN) trained to capture the system knowledge in a mostly auto-
mated manner. For instance, Figure 10.30 illustrates a simulation model where a
variable (i.e., predicted variable) is input to the model. That variable can be pre-
dicted using a DNN, which was trained to capture the new system state. In this
way, the DNN can be continuously trained in case of the arrival of new data that
echo new system states or conditions.

Study Limitations
A set of limitations should be acknowledged as follows:

◾ More big data-oriented scenarios can better present the potentials of integrat-
ing SM and ML.
◾ The clustering of patients was based on a mere data-driven perspective.
Adding a clinical viewpoint (e.g., diagnosis, procedures) may group patients
differently.
 Analytics Using Machine Learning-Guided Simulations ◾ 319

◾ All insights delivered by the study should be carefully considered as poten-

tial correlations limited by the dataset size and the number of variables
included.
◾ The IHFD dataset included records coming from public hospitals only.
◾ The rate of hip fractures was assumed to be constant over the simulated inter-
val; however, it might increase or decrease in reality.
◾ Due to lack of information, the study could not distinguish between patients
discharged to nursing homes or other residential care (e.g., rehabilitation
institutions).

Conclusions
The integration of simulation modeling and ML can help address further com-
plex questions and scenarios of analytics. We believe that the present work con-
tributes in this direction. The study can serve as a useful example of how ML
can assist the practice of modeling and simulation at different stages of model
development.
From a practical standpoint, it was also attempted to deliver useful insights in
relation to healthcare planning in Ireland, with a particular focus on hip fracture
care. The insights were provided based on a set of simulation models along with
ML predictions. At the population level, simulation models were used to mimic the
flow of patients, and the care journey, while ML provided accurate predictions of
care outcomes at the patient level.

References
Altmann, A., Toloşi, L., Sander, O., and Lengauer, T. (2010). Permutation importance:
A corrected feature importance measure. Bioinformatics, 26(10), 1340–1347.
Barga, R., Fontama, V., Tok, W. H., and Cabrera-Cordon, L. (2015). Predictive Analytics
with Microsoft Azure Machine Learning. New York: Apress.
Beyer, M. A., and Laney, D. (2012). The Importance of “Big Data”: A Definition. Stamford,
CT: Gartner, pp. 2014–2018.
Borshchev, A. (2013). The Big Book of Simulation Modeling: Multimethod Modeling with
AnyLogic 6. Chicago, IL: AnyLogic North America.
Brailsford, S. C., and Hilton, N. A. (2001). A comparison of discrete event simulation
and system dynamics for modelling health care systems. In Proceedings of ORAHS,
Glasgow, Scotland, pp. 18–39.
Brasel, K. J., Lim, H. J., Nirula, R., and Weigelt, J. A. (2007). Length of stay: An appropri-
ate quality measure? Archives of Surgery, 142(5), 461–466.
Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32.
British Orthopaedic Association. (2007). The Care of Patients with Fragility Fracture.
London, UK: British Orthopaedic Association, pp. 8–11.
320 ◾ Analytics and Knowledge Management

Cellier, F. E. (1991a). Qualitative modeling and simulation: Promise or illusion. In

Proceedings of the 23rd Conference on Winter Simulation. IEEE Computer Society,
pp. 1086–1090.
Cellier, F. E. (1991b). Continuous System Modeling. New York: Springer-Verlag.
Chawla, N. V., Bowyer, K. W., Hall, L. O., and Kegelmeyer, W. P. (2002). SMOTE: Synthetic
minority over-sampling technique. Journal of Artificial Intelligence Research, 16, 321–357.
Cleary, J., Goh, K. S., and Unger, B. W. (1985). Discrete event simulation in prolog. In
Proceedings of the Artificial Intelligence, Graphics and Simulation. San Diego, CA
Cohen, K. J., and Cyert, R. M. (1965). Simulation of Organizational Behavior. Handbook of
Organizations. Chicago, IL: Rand McNally, pp. 305–334.
Cooper, C., Campion, G., and Melton, L. 3. (1992). Hip fractures in the elderly: A world-
wide projection. Osteoporosis International, 2(6), 285–289.
CSO. (2017). Retrieved from http://www.cso.ie/en/statistics/population/.
Davenport, T. H., and Harris, J. G. (2007). Competing on Analytics: The New Science of
Winning. Boston, MA: Harvard Business Press.
De Mauro, A., Greco, M., and Grimaldi, M. (2015). What is big data? A consensual defini-
tion and a review of key research topics. In AIP Conference Proceedings (Vol. 1644,
No. 1). AIP, pp. 97–104.
Demchenko, Y., De Laat, C., and Membrey, P. (2014). Defining architecture components
of the big data ecosystem. In Proceedings of the 2014 International Conference on
Collaboration Technologies and Systems (CTS). IEEE, pp. 104–112.
Dijcks, J. P. (2012). Oracle: Big Data for the Enterprise. Oracle White Paper, Oracle Corporation.
Dodds, M. K., Codd, M. B., Looney, A., and Mulhall, K. J. (2009). Incidence of hip
fracture in the republic of Ireland and future projections: A population-based study.
Osteoporosis International, 20(12), 2105–2110.
Elbattah, M. (2017). Retrieved from https://github.com/Mahmoud-Elbattah/
DESModelImp.
Elbattah, M., and Molloy, O. (2016). Coupling simulation with machine learning: A
hybrid approach for elderly discharge planning. In Proceedings of the 2016 annual
ACM Conference on SIGSIM Principles of Advanced Discrete Simulation. New York:
ACM, pp. 47–56.
Ellanti, P., Cushen, B., Galbraith, A., Brent, L., Hurson, C., and Ahern, E. (2014).
Improving hip fracture care in Ireland: A preliminary report of the Irish hip fracture
database. Journal of Osteoporosis, 2014, 7.
Englert, J., Davis, K. M., and Koch, K. E. (2001). Using clinical practice analysis to improve
care. The Joint Commission Journal on Quality Improvement, 27(6), 291–301.
Faddy, M. J., and McClean, S. I. (1999). Analysing data on lengths of stay of hospital
patients using phase-type distributions. Applied Stochastic Models in Business and
Industry, 15(4), 311–317.
Fishwick, P. A. (1995). Computer Simulation: The Art and Science of Digital World
Construction. Gainesville, FL: Department of Computer and Information Science
and Engineering, University of Florida.
Fishwick, P. A., and Modjeski, R. B. (2012). Knowledge-Based Simulation: Methodology and
Application (Vol. 4). Berlin, Germany: Springer Science & Business Media.
Forrester, J.W. (1968). Principles of Systems. Cambridge, MA: MIT Press.
Fortune. (2012). What data says about us. Fortune, p. 163.
Futó, I., and Gergely, T. (1986). TS-PROLOG, a logic simulation language. Transactions of
the Society for Computer Simulation International, 3(4), 319–336.
 Analytics Using Machine Learning-Guided Simulations ◾ 321

Galar, M., Fernandez, A., Barrenechea, E., Bustince, H., and Herrera, F. (2012). A review
on ensembles for the class imbalance problem: Bagging-, boosting-, and hybrid-based
approaches. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications
and Reviews), 42(4), 463–484.
Gandomi, A., and Haider, M. (2015). Beyond the hype: Big data concepts, methods, and
analytics. International Journal of Information Management, 35(2), 137–144.
Harper, P. R., and Shahani, A. K. (2002). Modelling for the planning and management of
bed capacities in hospitals. Journal of the Operational Research Society, 53(1), 11–18.
HIPE. (2015). Retrieved from http://www.hpo.ie/hipe/hipe_data_dictionary/HIPE_
Data_Dictionary_2015_V7.0.pdf.
HSE. (2014a). Retrieved from http://www.hse.ie/eng/services/publications/olderpeople/
Executive_Summary_-_Strategy_to_Prevent_Falls_and_Fractures_in_Ireland%E2%
80%99s_Ageing_Population.pdf.
HSE. (2014b). Annual Report and Financial Statements 2014, Dublin, Ireland: Health
Service Executive (HSE).
Huang, Y. (2013). Automated simulation model generation. Doctoral dissertation, TU
Delft, Delft University of Technology.
Huang, Z. (1998). Extensions to the K-means algorithm for clustering large data sets with
categorical values. Data Mining and Knowledge Discovery, 2(3), 283–304.
IBM. (2017). Retrieved January 15, 2017, from http://www.ibmbigdatahub.com/
infographic/four-vs-big-data.
Intel. (2013). Peer Research report: Big data analytics. Retrieved February 10, 2017, from
http://www.intel.com/content/www/us/en/big.
Jain, A. K. (2010). Data clustering: 50 years beyond K-means. Pattern Recognition Letters,
31(8), 651–666.
Japkowicz, N., and Stephen, S. (2002). The class imbalance problem: A systematic study.
Intelligent Data Analysis, 6(5), 429–449.
Johansen, A., Wakeman, R., Boulton, C., Plant, F., Roberts, J., and Williams, A. (2013).
National Hip Fracture Database: National Report 2013. London, UK: Royal College
of Physicians.
Kuipers, B. (1986). Qualitative simulation. Artificial Intelligence, 29(3), 289–338.
Lane, D. C. (2000). You Just Don’t Understand Me: Modes of Failure and Success in the
Discourse Between System Dynamics and Discrete Event Simulation. London, UK: LSE
OR Department Working Paper LSEOR 00–34, London School of Economics and
Political Science.
Laney, D. (2001). 3D Data Management: Controlling Data Volume, Velocity and Variety.
Stamford, CT: META Group Research Note, 6, 70.
Lattner, A. D., Bogon, T., Lorion, Y., and Timm, I. J. (2010). A knowledge-based approach to
automated simulation model adaptation. In Proceedings of the 2010 Spring Simulation
Multiconference. Orlando, FL: Society for Computer Simulation International, p. 153.
Laursen, G. H., and Thorlund, J. (2016). Business Analytics for Managers: Taking Business
Intelligence beyond Reporting. Hoboken, NJ: John Wiley & Sons.
Law, A. M. (2008). How to build valid and credible simulation models. In Proceedings of the 40th
Conference on Winter Simulation. Miami, FL: Winter Simulation Conference, pp. 39–47.
LeCun, Y., Bengio, Y., and Hinton, G. (2015). Deep learning. Nature, 521(7553),
436–444.
Liberatore, M., and Luo, W. (2011). INFORMS and the analytics movement: The view of
the membership. Interfaces, 41(6), 578–589.
322 ◾ Analytics and Knowledge Management

Manyika, J., Chui, M., Brown, B., Bughin, J., Dobbs, R., Roxburgh, C., and Byers, A. H.
(2011). Big Data: The Next Frontier for Innovation, Competition, and Productivity.
Washington, DC: McKinsey Global Institute.
Maoz, M. (2013). How IT should deepen big data analysis to support customer-centricity.
Gartner G00248980.
Marshall, A. H., and McClean, S. I. (2003). Conditional phase-type distributions for
modelling patient length of stay in hospital. International Transactions in Operational
Research, 10(6), 565–576.
Martis, M. S. (2006). Validation of simulation based models: A theoretical outlook.
The Electronic Journal of Business Research Methods, 4(1), 39–46.
Mashey, J. R. (1997). Big data and the next wave of infraS-tress. In Computer Science
Division Seminar. Berkeley, CA: University of California.
Meadows, D. H. (2008). Thinking in Systems: A Primer. White River Junction, VT: Chelsea
Green Publishing.
Melton, L. J. (1996). Epidemiology of hip fractures: Implications of the exponential increase
with age. Bone, 18(3), S121–S125.
Microsoft. (2013). The big bang: How the big data wxplosion is changing the world—
Microsoft UK enterprise insights blog-site home-MSDN blogs. Retrieved February 2,
2017, from http://blogs.msdn.com/b/microsoftenterpriseinsight/archive/2013/04/15/
big-bang-how-the-big-data-explosion-is-changing-theworld.aspx.
Minelli, M., Chambers, M., and Dhiraj, A. (2012). Big Data, Big Analytics: Emerging Business
Intelligence and Analytic Trends for Today’s Businesses. New York: John Wiley & Sons.
Mortenson, M. J., Doherty, N. F., and Robinson, S. (2015). Operational research from
taylorism to terabytes: A research agenda for the analytics age. European Journal of
Operational Research, 241(3), 583–595.
Newell, A., and Simon, H. A. (1959). The Simulation of Human Thought. Current Trends in
Psychological Theory. Pittsburgh, PA: University of Pittsburgh Press.
NIST. (2015). S. 1500–1 NIST Big Data Interoperability Framework (NBDIF): Volume 1:
Definitions. Gaithersburg, MD.
NOCA. (2014). Irish hip fracture database national report 2014. National Office of Clinical
Audit (NOCA), Dublin, Ireland.
NOCA. (2017). Irish hip fracture database. Retrieved from https://www.noca.ie/
irish-hip-fracture-database.
O’Keefe, G. E., Jurkovich, G. J., and Maier, R. V. (1999). Defining excess resource utiliza-
tion and identifying associated factors for trauma victims. Journal of Trauma and
Acute Care Surgery, 46(3), 473–478.
Oberkampf, W. L., DeLand, S. M., Rutherford, B. M., Diegert, K. V., and Alvin, K. F.
(2002). Error and uncertainty in modeling and simulation. Reliability Engineering &
System Safety, 75(3), 333–357.
Oxford Dictionary. (2017). Retrieved from https://en.oxforddictionaries.com/definition/
big_data.
Patel, V. R., and Mehta, R. G. (2011). Impact of outlier removal and normalization approach
in modified K-Means clustering algorithm. IJCSI International Journal of Computer
Science Issues, 8(5), 331–336.
Powell, J., and Mustafee, N. (2014). Soft OR Approaches in problem formulation stage of
a hybrid M&S study. In Proceedings of the 2014 Winter Simulation Conference. IEEE
Press, pp. 1664–1675.
 Analytics Using Machine Learning-Guided Simulations ◾ 323

Rabelo, L., Cruz, L., Bhide, S., Joledo, O., Pastrana, J., and Xanthopoulos, P. (2014).
Analysis of the expansion of the panama canal using simulation modeling and artifi-
cial intelligence. In Proceedings of the 2014 Winter Simulation Conference. IEEE Press,
pp. 910–921.
Rashwan, W., Ragab, M., Abo-Hamad, W., and Arisha, A. (2013). Evaluating policy inter-
ventions for delayed discharge: A system dynamics approach. In Proceedings of the
2013 Winter Simulation Conference. IEEE Press, pp. 2463–2474.
Rindler, A., McLowry, S., and Hillard, R. (2013). Big Data Definition. MIKE2. 0, The
Open Source Methodology for Information Development. Retrieved from http://mike2.
openmethodology.org/wiki/Big_Data_Definition.
Rosenberg, M., and Everitt, J. (2001). Planning for aging populations: Inside or outside the
walls. Progress in Planning, 56(3), 119–168.
Shannon, R. E. (1975). Systems Simulation: The Art and Science. Englewood Cliffs, NJ:
Prentice-Hall.
Shreckengost, R. C. (1985). Dynamic simulation models: How valid are they? In Beatrice,
A. R., Nicholas, J. K., and Louise, G. R. (Eds.), Self-Report Methods of Estimating
Drug Use: Current Challenges to Validity. Rockville, MA: National Institute on Drug
Abuse Research Monograph, 57, pp. 63–70.
Soetaert, K. E. R., Petzoldt, T., and Setzer, R. W. (2010). Solving differential equations in
R: Package deSolve. Journal of Statistical Software, 33.
Stanislaw, H. (1986). Tests of computer simulation validity: What do they measure?
Simulation & Games, 17(2), 173–191.
Sterman, J. D. (1994). Learning in and about Complex Systems. System Dynamics Review,
10(2–3), 291–330.
Sun, Y., Wong, A. K., and Kamel, M. S. (2009). Classification of imbalanced data: A review.
International Journal of Pattern Recognition and Artificial Intelligence, 23(4), 687–719.
Sweetser, A. (1999). A comparison of system dynamics (SD) and discrete event simulation
(DES). In Proceedings of the 17th International Conference of the System Dynamics Society
Wellington, New Zealand., pp. 20–23.
Taylor, S. J., Khan, A., Morse, K. L., Tolk, A., Yilmaz, L., and Zander, J. (2013). Grand
challenges on the theory of modeling and simulation. In Proceedings of the Symposium
on Theory of Modeling & Simulation-DEVS Integrative M&S Symposium. San Diego,
CA: International Society for Computer Simulation.
Tolk, A. (2015). The next generation of modeling & simulation: Integrating big data and
deep learning. In Proceedings of the Conference on Summer Computer Simulation.
Chicago, IL: International Society for Computer Simulation.
Tolk, A., Balci, O., Combs, C. D., Fujimoto, R., Macal, C. M., Nelson, B. L., and Zimmerman,
P. (2015). Do we need a national research agenda for modeling and simulation?. In
Proceedings of the 2015 Winter Simulation Conference. IEEE Press, pp. 2571–2585.
UK Government. (2014). Retrieved January 4, 2017, from https://www.gov.uk/govern-
ment/uploads/system/uploads/attachment_data/file/389095/Horizon_Scanning_-_
Emerging_Technologies_Big_Data_report_1.pdf.
Visalakshi, N. K., and Thangavel, K. (2009). Impact of normalization in distributed
K-means clustering. International Journal of Soft Computing, 4(4), 168–172.
Ward, J. S., and Barker, A. (2013). Undefined by data: A survey of big data definitions.
arXiv preprint arXiv:1309.5821.
Wickham, H. (2009). ggplot2: Elegant Graphics for Data Analysis. New York: Springer
Science & Business Media.
324 ◾ Analytics and Knowledge Management

Wu, C., Buyya, R., and Ramamohanarao, K. (2016). Big data analytics= machine learn-
ing+ cloud computing. arXiv preprint arXiv:1601.03115.
Yang, Q., and Wu, X. (2006). 10 Challenging problems in data mining research.
International Journal of Information Technology & Decision Making, 5(4), 597–604.
Zhong, J., Cai, W., Luo, L., and Zhao, M. (2016). Learning behavior patterns from video
for agent-based crowd modeling and simulation. Autonomous Agents and Multi-Agent
Systems, 30(5), 990–1019.
Zikopoulos, P. C. (2013). Dirk deRoos, Krishnan Parasuraman, Thomas Deutsch, David
Corrigan, and James Giles. Harness the Power of Big Data. New York: The IBM Big
Data Platform.
Chapter 11

Intangible Dynamics:
Knowledge Assets in the
Context of Big Data and
Business Intelligence
G. Scott Erickson and Helen N. Rothberg

Contents
Introduction ......................................................................................................326
A Wider View of Intangibles .............................................................................326
Big Data and Business Analytics/Intelligence .....................................................328
Reimagining the Intangibles Hierarchy .............................................................329
Assessment of the Intelligence Hierarchy in Organizations ................................333
Measuring Intangible Asset Scenarios ................................................................336
Intangible Assets and Metrics: Illustrative Applications..................................... 340
Healthcare ............................................................................................... 340
Financial Services..................................................................................... 344
Automated Driving ..................................................................................347
Conclusions ...................................................................................................... 351
References .........................................................................................................352

325
326 ◾ Analytics and Knowledge Management

Introduction
The explosive growth of interest in Big Data and analytics has caught the attention
of knowledge management (KM) researchers and practitioners. While some over-
laps are clear between the fields, previous interest in data or information from the
KM field was rather limited. Over the development of the KM discipline, highly
valued knowledge was explicitly differentiated from purportedly uninteresting
data. That perspective is changing.
This chapter will take an updated look at the nature of intangible assets, not
just knowledge but related intangibles including data, information, and wisdom or
intelligence. In clarifying the similarities and differences, we believe both practitio-
ners and scholars can more effectively understand these potentially valuable assets,
providing a greater opportunity to exploit them for competitive advantage. A struc-
tured understanding of all intangible assets, and their inter-relationships, can lead
to better applications and results.
After a review of existing theory concerning types of intangibles and approaches
for exploiting or managing them, we will turn to some data-driven examples of
how a fuller understanding of the combined disciplines can help with strategic
decision-making. In particular, we will consider how a firm’s intangibles compe-
tencies translate into success in its industry sector as well as other directions for
potential growth.

A Wider View of Intangibles

Even before the advent of KM and intellectual capital (IC) as well-established dis-
ciplines, an understanding existed of the differences between data and knowledge.
In particular, Ackoff’s (1989) data-information-knowledge-wisdom (DIKW) hier-
archy has often been referred to as the basis for much of our discussion concerning
all intangible assets, defining these as progressing from data and information to
knowledge and wisdom. As KM developed in the 1990s, early scholarship often
referred to these definitions for context. The field used the hierarchy to firmly
entrench exactly what is a valuable intangible (knowledge, gained from learned
know-how, placing information in context) versus what is not (Zack, 1999). The
latter includes data (unorganized observations), information (categorized data), and
wisdom (generally too personal to manage as knowledge). IC writings made some
similar distinctions, often referring to knowledge as the key to IC but sometimes
including some of the other intangibles in wider descriptions (Stewart, 1997).
As a result, much of the attention focused on knowledge had to do with defin-
ing its characteristics and exploiting it to its best advantage. Interestingly, the
creation of knowledge, though acknowledged, did not receive as much attention
as sharing the knowledge once it exists and is recognized. Part of this is due to
the original insight from Nonaka and Takeuchi (1995) that individuals learn, not
 Intangible Dynamics ◾ 327

organizations. Knowledge comes from individuals learning from experience, gain-

ing hands-on know-how. That can be hard for the organization to manage, though
it can create the right conditions for learning, support learning, and provide oppor-
tunities to employ new learning. Even so, the study of innovation and from where
new processes and new products come is generally seen as a separate field. While
acknowledging that the knowledge needs to be created, almost exogenously, KM
has focused heavily on what to do with it afterwards. The emphasis is on exchange
and sharing, and most of what we study is centered on how to do that effectively.
If one considers the main streams of the literature over the years, scholars and
practitioners have tended to try to understand the nature of knowledge assets, their
different types, and then what to do with them based on that nature. On the KM
side, the attention usually starts with the tacit or explicit distinction (Nonaka and
Takeuchi, 1995; Polanyi, 1967); how explicit knowledge can be more easily codi-
fied, captured, and distributed by the organization while tacit knowledge is more
personal and harder to communicate. From this difference, the field has developed
approaches to better share the knowledge, from IT-based tools for more explicit
knowledge to person-to-person tools for more tacit knowledge (and variations in
between) (Brown and Duguid, 1991; Matson et al., 2003; Thomas et al., 2001).
Other variables concerning the knowledge itself, from complexity to stickiness
can also matter in how it is managed (Kogut and Zander, 1992; McEvily and
Chakravarthy, 2002; Zander and Kogut, 1995). Similarly, variations across organi-
zations such as social capital, social networks, and so forth can also be a factor in
how firms plan for and use knowledge (Liebowitz, 2005; Nahapiet and Ghoshal,
1998).
On the IC side, we have seen similar things. Here, although tacit or explicit is
also recognized, the distinction between human (job-related knowledge), structural
(organization-related knowledge), and relational (knowledge concerning external
parties such as customers) capital is also important (Bontis, 1999; Edvinsson and
Malone, 1997; Stewart, 1997). Similar to KM, the idea is that when organiza-
tions better understand the nature of their assets, they can better manage them.
Consequently, knowledge audits can be conducted (Marr et al., 2004) and deeper
dives into the details of the IC pieces (e.g., human capital) can be done to better
assess and then better exploit these knowledge assets. Increasingly specific differ-
ences between human capital have become a subject of some extensive interest in
some scholarly circles, especially as regards unique individuals or unique talents
(“stars”) that might require individual attention (Crocker, 2014; Groysberg et al.,
2010).
All of this can be important if one buys into the notion of intangibles having an
impact on the competitive potential of organizations. Flowing from the resource-
based view of the firm positing that an organization must possess some unique,
sustainable differentiator to be successful (Wernerfelt, 1984), researchers in the
field believe intangible assets such as knowledge or IC can be that unique resource
(Grant, 1996; Teece, 1998). There is a school of thought that intangibles are the
328 ◾ Analytics and Knowledge Management

only really, truly unique resource to be found given the ubiquity of basic labor, capi-
tal, information technology (IT), and so on, in today’s world. The only unique thing
most firms having going for them is to be found in the heads of their key employees.
Consequently, the fields of KM and IC are fairly established in terms of defin-
ing the nature of knowledge assets, how they vary, and what to do with them. But
over the past few years, the rapid growth of interest in big data applications has led
to questions of how such matters relate to knowledge, particularly KM that has
been dismissive of the value of “less-developed” data and information. Is there a
means to reconcile the disciplines and include the full range of intangible assets in
our discussions?

Big Data and Business Analytics/Intelligence

One of the important things to understand about Big Data and business analytics is
the lack of understanding among much of the business community about what the
terms actually mean. For those coming from a KM or IC perspective, the defini-
tions are actually straightforward even if there is a general lack of appreciation for
the potential of the fields. Big Data refers to the stream of bytes generated by infor-
mation systems observing behavior, communications, or other activities. When
organized, it can become information. The phenomenon of Big Data has to do with
the exploding amounts of data suddenly available, characterized in the field by the
“3 Vs” of volume, variety, and velocity (Laney, 2001). Ever increasing amounts of
data (volume) in an expanding number of forms, including unstructured data such
as text, images, and video (variety), are being generated, transmitted, and received
ever more quickly (velocity) (Chen et al., 2012; McAfee and Brynjolfsson, 2012).
The plunging costs of storing and processing such data are feeding these trends and
are the reason so many participants are excited about the cloud—the manifestation
of cheap data handling.
But this first stage of Big Data generally limits itself to collection, organiza-
tion, and transmission of the data. There is exchange, and there can be action, but
the data don’t require additional analysis to be useful. Big Data, in and of itself,
simply tracks observations of interest. It is exchanged across a company or across
its extended network, but often just reported on dashboards or similar devices.
Key performance indicators (KPIs) might be tracked and given special attention.
The KPIs may even have tolerances attached to them, much the same as quality
management systems track performance data to ensure adherence to standards.
Something out of tolerance or with an abnormal pattern to the data might be
further investigated or acted upon. Algorithms may be established so that the
system automatically makes a decision to intervene. The types of data collected
are often operational, transactional, or communications (especially digital media),
but anything generating large amounts of data can be included in these big data
platforms.
 Intangible Dynamics ◾ 329

What is important to understand, however, is that these systems generally only

organize and transfer the data. Someone or something may monitor the results but the
data are not fundamentally changed in any way. Further analysis on the available data
is not really a part of these systems. That capability is the interesting part of the second
generation of Big Data, big data analytics. Also referred to as business intelligence,
the analytics function subjects Big Data to analysis leading to new insights, effectively
learning something new from studying the data. Specialized techniques, enabled
by cutting-edge analysis software, are available for this type of work characterized by
data mining, predictive analytics, clustering, and other approaches. Typically, analysts
are doing a deep dive into the data looking for nonobvious correlations or differences.
The new insights or learning require highly skilled individuals using advanced data
processing and statistical analysis tools, though there is also some potential in artificial
intelligence to uncover and test new insights in such environments.
Big Data and business analytics are major and connected advances in the use of
intangible assets to better steer and run organizations but, as just explained, they
are two different things. Big data systems collect and transfer observations, chiefly
monitoring them and acting on deviations. Big data analytics organizes and sub-
jects the data to deep analysis with advanced methods, looking for new insights.
And in the context of how we see intangibles, how we define them, and how we
make recommendations for applying them, those differences can be important.
Important enough to reconsider how we characterize the entire field of intangible
assets even if those in KM want to continue to refer to them as knowledge or as IC.

Reimagining the Intangibles Hierarchy

Earlier in this chapter, we mentioned the traditional DIKW hierarchy, the pro-
gression from data through information, knowledge, and wisdom. While this has
had its uses over the years, it has also fallen into some disuse in the KM and IC
fields specifically because they focus so precisely on just the knowledge component.
Moreover, the advances in the fields of Big Data and business analytics, while hav-
ing some clear connections to parts of the hierarchy (data) are less well-explained
by the entire structure. If we are to incorporate both traditional KM and IC assets
and the new potential we see from Big Data and from business analytics and intel-
ligence, a different framework might have a better fit.
One possibility flows from Kurtz and Snowden’s (2003) Cynefin sense-making
framework. Though not necessarily intended in the manner of an intangibles
hierarchy, it can and has been applied in such a manner (Rothberg and Erickson,
2017; Simard, 2014). The driver of the framework is understanding and process-
ing different environments, from ordered to chaotic, but the real contribution is in
describing those situations in a context related to identifying and collecting intan-
gible inputs, making sense of them, and acting appropriately. In that sense, the best
place to start might be with the dimensions of Cynefin.
330 ◾ Analytics and Knowledge Management

Kurtz and Snowden divide the world into sectors based on the dimensions of
centralization and hierarchy versus connections and networks (“meshwork” in one
extension). In this conceptualization, do intangibles flow into the center (poten-
tially redistributable), across individuals without going to the center, both, or nei-
ther? The different scenarios both help to explain how intangibles might be best
understood and used most effectively. The basic ideas, including their potential
contributions and applications, can be explained as follows:

◾ Known: This domain is defined by centralization and boundary network

connections. It is ordered, with observable, notable patterns. The intangibles
are easily shared across all levels and locations and, if captured by the cen-
tral core, easily dispersed back out into the network. This pattern fits what
organizations do with data and information. It is only bytes, easily trans-
ferred from individual to individual or from the data-gathering point to the
database at the center of the firm. Patterns are monitored, divergence from
expected trends is noted and acted upon, and decisions are generally based
on quick action, pattern recognition, and operational outcomes more than
long-term strategic planning.
Kurtz (2009) characterizes the known domain similar to a clock mecha-
nism: complicated but with observable, understandable repeated patterns.
Verdun (2008) suggests that a traditional control hierarchy is appropriate for
managing such circumstances as the operation is orchestrated through cen-
tral direction and is focused on individual jobs and places. That being said, if
the same environment recurs throughout the organization (multiple plants,
multiple sales offices), data or information about the processes is certainly
useful to centralize or just redistribute across individuals or locations. Going
back to Kurtz, the scenario would have both central connections (feeding the
center) and constituent connections (across the periphery).
◾ Knowable: The knowable domain shows centralization but fewer connections
to other networks. It is still ordered, but the patterns are less obvious or yet
unknown, and require some experience to recognize and learn from. Once
knowledge is created, it can be codified and shared with the center, again
available for dispersion back out across the wider network. When appropri-
ate, other individuals may see opportunities to apply the knowledge to their
own situations. Patterns are noticed and individuals learn from them, though
the decisions largely center on operations and customer relationships, day-to-
day effectiveness, more than long-term insights. This scenario better fits what
organizations do with explicit knowledge.
Verdon describes this as a blend of orchestrated and adaptive, with indi-
viduals given structure by the center but also learning things on their own as
their circumstances dictate. Any new knowledge is fed back to the center from
the periphery, then distributed back out to others in similar environments
who may be able to use it. But it is not a center-driven knowledge exchange;
 Intangible Dynamics ◾ 331

it depends on contributions from outside. Hence, a “modular heterarchical”

structure connecting the periphery with the center and with communications
running in both directions makes more sense. Kurtz characterizes knowable
as having prominent central connections but less distinct constituent ones.
◾ Complex: Complexity is a domain without centralization but with periph-
eral network connections. Order is reduced, fewer patterns are apparent.
Knowledge is once again created by individuals but is more difficult to com-
municate. As a result, centralization or capture by the core organization is
also difficult. But person-to-person knowledge exchange throughout the
extended network is quite possible, so one sees the application of more tacit
knowledge-related techniques such as communities of practice, mentoring,
apprenticeships, and so on, being applied. Decisions largely remain about
operational matters but now depend more on the level of the individuals
doing the exchange. If at the senior manager level, they may well be about
how to do strategic planning better, so operational decisions are still pre-
eminent but strategic and tactical decisions now begin to enter the conversation.
As noted earlier, tacit knowledge fits this domain.
Kurtz describes the complex domain as being unordered and with pat-
terns but patterns that can unpredictably change. These may be simple pat-
terns, discernible but based on large numbers of interactions or observations.
Once individuals develop some knowledge, it is difficult to communicate that
understanding to someone else without some context as contingencies, abil-
ity to anticipate pattern changes, and so forth make the knowledge harder to
transfer. Verdun recommends a “customized—responsible autonomy” struc-
ture for management, providing individuals with the freedom to act given
their own circumstances and the resulting ability to learn from their actions.
Connections can be made wherever they are useful, across individual partici-
pants anywhere in the organization, so the context is person-centric and agile
and adaptable. Kurtz’ representation would be more constituent-connected
and less central-connected. It is in this environment that Brown and Boudes
(2005), commenting on the Cynefin approach, note the beginnings of its
emphasis on a diversity of opinions, action outside the bureaucracy, and a
willingness to accept complexity rather than standardized solutions. Problem
solving in these environments starts to come from non-linear thinking and
different perspectives.
◾ Chaos: Chaos represents a lack of centralization and a lack of networking.
Order and patterns are gone. Insights are created by unique individuals but
they may find it difficult or even impossible to share their creative processes
with others. If one is familiar with Snowden’s work, this is a scenario where
subconscious communication might be used (storytelling) but even that can
be difficult. The learning is unique to the individual and often stays with
them. Since such rare individuals are often valuable to the entire enterprise,
now one sees the level of decision-making rise, and longer-term strategic,
332 ◾ Analytics and Knowledge Management

tactical, or innovative insights are more common. This environment not only
fits C-suite executives but also the big data analysts discovering unexpected
insights buried in the data lakes available to data-heavy organizations. Much
like R&D innovators or charismatic leaders, what they do may be difficult
to understand, let alone teach someone else to do. The results of their efforts
might be perfectly understandable but how they get to that is opaque and
potentially unlearnable. This domain characterizes intelligence, the ability to
draw new insights by analyzing a variety of intangible assets.
This domain comes full circle to a non-bureaucratic approach to learn-
ing and creating intangibles of value to the organization. Kurtz now sug-
gests intricate patterns (still liable to unexpectedly change) based on simple
operations. She uses the example of a government project looking to identify
weak signals in public information and news reports to predict a future envi-
ronment, specifically noting how hard it was to teach participants to think
outside their normal frames of reference. In chaos, it is the unexpected insight
or spark of creativity that is critical, the different perspective, and while that
solution might be communicable or teachable, the process to get to it might
not be. Verdun uses the term improvisation and does not really recommend
any structure, suggesting any and all connections may be useful. Kurtz would
note the lack of both central connections and constituent connections. As
just noted, this is where Snowden’s emphasis on stories instead of rational lin-
guistic explanation make sense as the learning, if possible, might take place
subconsciously rather than consciously. And, once again, Brown and Boudes’
emphasis on diverse perspectives, outside-the-network individual thinking,
and non-conformity would resonate for organizations trying to operate and
create or manage intangibles in such environments.

Following the logic, a new hierarchy evolves (Rothberg and Erickson, 2017; Simard,
2014). This time, the flow starts with data and information, moves through explicit
knowledge, tacit knowledge, and on to insight and intelligence. Per Kurtz’ (2009)
own commentary on the framework, this hierarchy does not imply differences in
value, the higher levels of the hierarchy are not necessarily better or worse than
lower levels. But there are differences in environments, and the hierarchy helps
with understanding how to manage under the different circumstances. Given the
new circumstances engendered by Big Data and business analytics and intelligence,
included now at opposite ends of this intangible asset hierarchy, we believe this
structure helps to both explain the nature of the different environments with differ-
ent intangibles present while also suggesting guidelines for handling them.
If the system is only designed to manage Big Data (“known” scenario), IT han-
dling the exchange of data and information is enough. As noted, this may include
dashboards to track metrics of high interest or established algorithms to react to
results not at expected levels. But other than the way they are organized, these
intangibles do not really require further analysis or more learning to have an impact.
 Intangible Dynamics ◾ 333

If explicit knowledge is critical to the firm or within its industry (“knowable”

scenario), then job-specific individual learning takes place that improves internal
processes, external relationships, or other aspects of the organization. The appropri-
ate system for managing such knowledge assets can be IT-based so as to leverage
them through further distribution though decision makers. These are the KM sys-
tems that have been discussed in the literature for decades and are likely familiar to
those with experience in the field.
If tacit knowledge is prominent (“complex” scenarios), individuals again learn
and develop new knowledge but it is of a more personal nature and more difficult to
communicate to others, especially in digital form. It can be done (e.g., case histories
or best practices stored in IT systems) but is often more effective if transmitted on a
person-to-person basis as with communities of practice, mentoring and apprentice-
ships, storytelling, and so forth. Again, these tacit KM approaches are well-known
in the discipline; established tools for transferring more complex knowledge requir-
ing context or some other deeper explanation.
Finally, if intelligence is evident in the firm or its industry (“chaos” scenar-
ios), then a system should be available to more effectively manage the process. As
noted in the discussion earlier, the path to individual insight and creativity may
be unique and extremely hard, if not impossible to communicate to others. The
intelligence just comes from the unique individual who can analyze intangible
inputs of all types and find something new in the chaos. On the other hand, the
literature does provide some guidance in how firms can structure themselves to
encourage such individual brilliance, specifically learning organizations designed
to feed inputs to individuals or diverse teams charged with finding insight by
applying their different perspectives (Argyris, 1992, 1993, 1977; Senge, 1990).
In another setting, we have called such structures intelligent learning organiza-
tions (Rothberg and Erickson, 2017), and they provide a targeted analytics process
intended precisely at generating intelligence in a way Big Data or KM systems,
focused only on exchange, do not.

Assessment of the Intelligence Hierarchy

in Organizations
Much of this thought and the methodology behind it has been developed
much more extensively in other research (Erickson and Rothberg, 2012, 2016;
Rothberg and Erickson, 2017) and the intent is not to repeat that here. What
is useful to note is that different sorts of intangibles exist, they are successfully
applied in different types of circumstances, and so a strategic approach to iden-
tifying and developing the right intangibles to enhance a firm’s competitiveness
is required (Rothberg and Erickson, 2005). And we have a structured way to
analyze those circumstances as well as appropriate metrics for identifying and
then understanding them.
334 ◾ Analytics and Knowledge Management

◾ High levels of Big Data, knowledge, and intelligence suggest that all of the
intangibles are present and important. Industries with these sorts of metrics
include pharmaceuticals, software, and similar environments that not only
require substantial human capital but also efficient and high-quality pro-
cesses (structural capital) and effective, data- and relationship-driven mar-
keting (relational capital). Pharmaceuticals, for example, employs Big Data
in its operations, distribution, and marketing and sales; explicit knowledge in
its operational processes and sales; but more tacit knowledge and intelligence
in its labs and in analytics explorations to improve more strategic and tactical
approaches throughout the firm (R&D, production, distribution channels,
marketing, and consumer experience).
◾ High levels of Big Data and evidence of intelligence but low knowledge
metrics suggest that Big Data and analytics and insight (perhaps some tacit
knowledge bleeding in as well) are important, but explicit knowledge is not.
Big Data is available and value can come from conducting deeper analysis (data
mining, predictive analytics) on it. But the ability to discover such insights is
rare and difficult to share with others. Operations, transactions, and market-
ing are routine (structural capital and relational capital) and human capital is
found in the rare analyst or team who can find something new in the data.
Financial services are a typical industry, awash in data and aggressive competi-
tive intelligence but very low knowledge metrics compared to other industries.
◾ Moderate to high levels of big data and high knowledge metrics but little
intelligence activity suggest that data and knowledge are involved in scale and
efficiency, making processes of all kinds run better but that there is little in
the way of new insights or innovation. Particularly notable is that we believe
these types of industries have considerable explicit knowledge but probably
less tacit. What is known can be and is widely shared. Human capital (supply
chain, operations, and transactions), structural capital (in the same areas),
and relational capital (customer relationships) are all important but gener-
ally operational rather than strategic. Characteristic industries are consumer
goods and retail. Big brands (high relational capital) and operational size and
efficiency (human and structural capital) are key competitive elements.
◾ Moderate to low levels of big data, low knowledge, and low intelligence
(very little of anything of value). What data and knowledge (explicit) there
is should be supported, but operational, transactional, and marketing activi-
ties are well-understood and have little to be discovered that might improve
them. Much of the IC is probably structural or relational, investing in high
human capital employees never hurts but does not necessarily generate a great
return either, so cost and benefit should be considered. These are usually
mature manufacturing industries or regulated industries like utilities.

Again, the main idea in this chapter is not to recreate the existing logic and evi-
dence behind these distinctions but to introduce them as evidence of the wide range
 Intangible Dynamics ◾ 335

of circumstances facing decision makers thinking about how to manage all these
intangibles. Intangible dynamics are situational and call for different approaches.
Circumstances may even call for different metrics and some of the all-encompassing
measurement systems discussed in the literature may need to be broken back down
to lower levels to really help. But, as a first pass, we already have some established
metrics that allow an evaluation of different industries and individual firms’ com-
petitiveness within them.
Big Data can be measured directly by looking at relative data holdings in dif-
ferent industries. One often-cited report was done by McKinsey Global Services
(Manyika et al., 2011). The industry categories are fairly broad but the data provide
a good idea of where Big Data is prevalent (financial services, communications and
media, discrete and process manufacturing, utilities) and where it is not (construc-
tion, consumer and recreation services, professional services, healthcare providers).
Moreover, the study adds some details about the nature of the data and how it
might be used in these industries.
KM metrics, of course, remains an area of considerable discourse and lack of
consensus (Firer and Williams, 2003; Tan et al., 2007). Sveiby (2010) noted over
40 different approaches that have been applied by researchers or practitioners, spe-
cifically categorizing them by level of measurement (entire organization or indi-
viduals) and method (financial statements to survey instruments). In this chapter,
as we try to evaluate multiple firms in multiple industries; the low-level, individual
employee approaches do not really work. What has been used and what works
in cross-industry comparisons are objective financial results, even if there can be
issues with trying to apply them too precisely. Specifically, a modified Tobin’s q can
be used (market capitalization to book value or market capitalization to assets) to
make broad distinctions between industries and even individual firms (Erickson
and Rothberg, 2012; Tobin and Brainard, 1977). As noted above, however, we’ve
come to the conclusion that a high Tobin’s q, all by itself, probably only identifies
explicit knowledge. Tacit knowledge does not scale enough to show up reliably in
such metrics though it may also be present. But we can employ another metric that
might flag it when used in combination with Tobin’s q.
Intelligence, as defined in this chapter, signifies organizations seeking unique,
creative insights from analytical techniques. Again, this may be highly individual
and impossible to teach. Or it may be from a form of intelligent learning organiza-
tion established to funnel inputs and provide room for learning and interchange
between dedicated analysts with different perspectives. This capability can be
particularly hard to identify, but we’ve found that organizations with an intelli-
gence capacity in one area (e.g., R&D laboratories, competitive intelligence) know
something about how to do it and so can extend the capability to other areas.
So, again, an established metric exists of identifying competitive intelligence (CI)
operations in organizations as a proxy for the ability to establish an analytics
capability, drawing new insights or creativity from intangibles inputs (Erickson
and Rothberg, 2012).
336 ◾ Analytics and Knowledge Management

Measuring Intangible Asset Scenarios

The specific empirical data included in this chapter follows these general descrip-
tions and, as a whole, can be used to identify, quantify, and explain the intangibles
capabilities and related domains in different industry sectors. As characterized by
the hierarchy and the prominent intangibles characteristics present, we have a foun-
dation for assessing the competitive conditions in each sector. Once we possess a
broad idea of what intangibles are typically present and necessary for success, we
can also form an idea of the intangibles management capabilities are required to
compete (especially for new entrants). And with deeper analysis, we can start to
understand the actual activities, along the value chain, for example, that are related
to the intangibles. We’ll cover these shortly.
Regarding intangible assets such as data and information, the amount of data
storage per industry is a good start to identifying the amount of useful data gener-
ated and kept. As noted, the McKinsey study (Manyika et al., 2011) which reports
such numbers, has been very useful for a number of studies specifically on this
question. Even though the study is starting to show its age, there’s no reason to
believe that the numbers, at least on a relative basis, have changed appreciably since
publication. For our purposes, and even more interesting for assessing what firms
need in particular sectors, we can employ the report’s breakdown of data storage in
each industry by firm and discover the average big data holding on a company-by-
company basis. Here, we employ McKinsey’s metrics on data holdings (terabytes)
per industry per firm.
What large holdings of data indicate are systems capable of generating big
data and some sense of value seen in processing it and holding it. As indicated in
Table 11.1, the generation is often by machine, observing an activity or behavior,
then feeding the observations into the data system. So typical areas generating such
data would include supply chain logistics, operations (goods production, service
delivery), transactions (retail, financial), communications (direct to customer, social
media monitoring), and similar high-volume environments. As also noted, the sys-
tems used only to manage Big Data typically do no more than collect, process, and
report the results on an ongoing basis. Given KPIs, established tolerances, perfor-
mance levels, or other “triggers” will demand attention from a decision maker or an
embedded algorithm will initiate a planned reaction. The presence of high big data
holdings (high data storage per firm) and only high big data holdings indicates that
only the big data system is needed and employed (a “known” sense-making envi-
ronment). If high big data holdings are present, along with other intangibles, Big
Data contributes to those, and one of the other sense-making environments exists.
As indicated earlier, to assess intangible assets beyond data, such as knowledge,
we can employ a modified Tobin’s q. The value of this metric is that it can be easily
calculated and compared across firms, really the only way to do a cross-industry
study such as this with large numbers of per firm observations. Reported here are
two variations on the Tobin’s q, market capitalization to book value and market
 Intangible Dynamics ◾ 337

Table 11.1 Intangibles, Metrics, and Sense-Making Environments

Explicit Insight/
Data/Information Knowledge Tacit Knowledge Intelligence

Domain Known Knowable Complex Chaos

Definition Collected Learned Learned Nonobvious

observations, know-how, know-how, insights from
bytes codifiable, personalized, individual
sharable hard to codify brilliance or
or share applied
analysis
Source Data collectors, Individuals, Individuals, Individual or
often machines learning over learning over team,
time time insightful
discovery,
often from
analysis
Exchange IT systems IT systems or Personal Personal
personal interaction or interaction, if
interaction IT systems at all
Range Extended Organization Groups Individual
organizational
network
Management Big data systems Knowledge Knowledge Intelligence
management management systems,
systems systems learning
(IT-centered) (person-to- systems
person
centered)
Metrics Data storage per Modified Modified Tobin’s CI activity
firm Tobin’s q q and CI
activity
Indicates If only data If only a high High Tobin’s q High CI
storage per firm, Tobin’s q, only and high CI activity, if
only Big Data is explicit indicate tacit only with
present. knowledge is knowledge is high data
present. present (along storage,
with other indicates
intangibles). business
analytics
present. If by
itself or with
other
variables,
other
intelligence
is present.
338 ◾ Analytics and Knowledge Management

capitalization to asset value. We’ve tested both variations in multiple circumstances

(Erickson and Rothberg, 2012), including specifically to evaluate whether critical
differences are apparent (Erickson and Rothberg, 2013b). Where differences occur,
they are often in circumstances where liabilities are considerable. The key difference
in the metrics is performance given ownership of the assets (cap/book, where book
is assets less liabilities) versus performance given asset levels regardless of owner-
ship (cap/assets). Generally, we’ve found the two metrics agree, and one can push
ahead with analysis with some confidence when that occurs. It is of interest and
worth discussion when they do not (generally cap/assets is the more reliable one),
and researchers would want to look more deeply into the reasons behind the result.
Here, both are reported. We have created two databases, one including financial
results from 2005 to 2009 for all firms reporting on North American exchanges,
the other for the same conditions but covering 2010–2014. The first database
includes almost 2,000 firms and over 7,000 observations. The second is still under
construction, some targeted results were pulled for this chapter, but looks as if it
will generate similar descriptors from similar numbers of firms and observations.
From the full databases, only yearly observations with firms generating $1 billion
in revenue were included, as were industry sectors with at least twenty observations
(sectors were sometimes expanded if tangential sectors seemed complementary).
High ratings on the Tobin’s q variations (benchmarked against the database
means) indicate significant intangible assets, interpreted here and elsewhere as
knowledge, and an ability to turn a given amount of tangible assets into value for
the firm. More value relative to assets suggests a better job in creating and managing
intangibles. But it is relative to an industry or industry sector. The level of tangible
assets and what intangibles can add to them varies dramatically by circumstance.
Further, as discussed earlier, the skill in managing the knowledge-related intangible
assets must be impactful enough to drive market capitalization, the perceived value
of the firm. Tacit knowledge in a few heads isn’t going to register in billion dollar
corporations except in very unique circumstances. But explicit knowledge, scaled
up and applied usefully throughout the firm might be. Specifically, skillfully-
applied explicit knowledge can result in more efficient supply chains, operations, or
marketing relationships (including brand equity). That shows up in the Tobin’s q
metric. So a high value on this measure is indicative of substantial explicit knowl-
edge, all other things being equal, especially if paired with high Big Data stocks.
The two together can drive efficiency and indicate productive operations and effec-
tive marketing, even without a high intelligence value associated with the firm.
That intangible asset characterized as intelligence can be observed in a metric
used to assess competitive intelligence in firms. Here, we can apply data harvested
from continuous surveys done by CI consultancy Fuld & Company across the years
2005–2009 (matching the terms of the McKinsey data and the first Tobin’s q data-
set). A later dataset is again in the process of construction but no concrete results are
yet available. Early indications are that relative levels of CI activity have remained
consistent over time.
 Intangible Dynamics ◾ 339

This particular dataset is based on two pieces of data used to construct an index.
The number of participants from firms within an industry sectors is one input. An
industry sector with eighteen survey respondents is a quite different environment
than one with one or none. Further, each survey asks respondents to report on the
professionalism of their CI operation, from just starting to highly proficient. Again,
this indicator shows considerable differences between inexperienced, perhaps one-
person initiatives versus large, seasoned teams, even if only one respondent was
included from each firm. Combined, the index presented gives a sense of both
quantity and quality of CI efforts in a given industry sector. As discussed earlier, if
a firm is capable of intelligence activity in one area (competitive intelligence), it has
the ability to practice it in others (marketing intelligence and analytics, business
intelligence and analytics).
Interpretation of the indicator, as with others, depends on circumstances and
what other intangibles seem to be present. Just the intelligence metric, with no
other intangibles (very rarely seen) or with Big Data (common in identifiable sec-
tors), indicates an insight or analytic capability. Firms in these sectors are able to
create new insights, particularly when analyzing big data inputs. Such capabilities
are often unique to a particularly gifted individual or team within the firm and,
unlike knowledge assets, can be extremely difficult if not impossible to teach to
others. The results of the insight are replicable but the process to get there is associ-
ated only with the individual or team generating it. Such circumstances are seen
when firms have gifted leaders or creative talent, the situations when “stars” are
uniquely valuable and need to be managed as a rare asset in and of themselves.
Alternatively, when the intelligence indicator is present with a high Tobin’s q
or knowledge metric, tacit knowledge is also present. Again, unique insight comes
from individuals, as indicated by the intelligence result. But at least some of these
learnings occur more often and are more scalable—they occur in circumstances
that make them scalable. That doesn’t mean intelligence itself isn’t available as well,
just that tacit and explicit knowledge are apparent, too. Pharmaceutical companies,
for example, typically have all the intangibles: Big Data, both types of knowledge,
and intelligence. They have huge amounts of data from operations, their distribution
channels, and their customers (providers, retailers, insurers, and end consumers).
They also have explicit and tacit knowledge on how to make their processes better,
their marketing relationships better, and their labs run better. But they still need
individual insight as well, in the creativity and brilliance of their researchers as well
as for occasional strategic concepts or new marketing directions.
In the end, we can identify the different circumstances where different intan-
gibles are more or less prominent in an industry sector. We can spot industries
using only Big Data, transferring and monitoring it to support operational and
marketing efficiency. We can spot industries where explicit knowledge is critical,
where employees learn to make improvements in logistics, operations, marketing,
and any number of other functions of the firm. These learnings can then be scaled
up for greater impact through appropriate KM systems. We can spot industries
340 ◾ Analytics and Knowledge Management

where tacit knowledge and intelligence are also present, either by themselves (more
intelligence-oriented) or with explicit knowledge (indicating both tacit knowledge
and intelligence).
As a consequence, decision makers gain a better appreciation of what intangi-
bles contribute to competitive advantage in such sectors and how they might stack
up against rival firms. The indicators also provide guidance as to where to focus
attention in order to take a deeper look. So explicit knowledge might be indicated,
for example, but what is the nature of it? Is it in manufacturing goods or service
delivery? Is it in distribution efficiency? Is it in close customer relationships or deep
customer knowledge? By knowing where to look and having a rough idea what to
look for, strategists stand a better chance of truly understanding what intangibles
are necessary and how they need to be managed.

Intangible Assets and Metrics: Illustrative Applications

To demonstrate, we’ll look at a few applications, including pertinent data indicating
different intangibles scenarios. In some ways, the conclusions might seem obvious,
but we picked distinctive examples allowing clear discussion even at a fairly general
level. The purpose here is demonstration of process, an actual strategic application
might show less differences between sector values and would require considerably
more depth in the follow-on data and information collection. A deeper dive could
lead to a better understanding of what the difference in intangible assets mean to
competitiveness.
The scenarios all revolve around insurance firms. In part, this is because we
have profit pools already established in past studies by Bain (Eliades et al., 2012;
Schwedel and Rodrigues, 2009) and don’t need to construct new ones just for this
chapter. The healthcare profit pool is included as is the financial services profit pool,
both of which include insurance sectors. To this, we add a more general discussion
on the free-for-all in the automobile industry as demonstrated by reported interest
in self-driving cars. Although not a profit pool yet, this new direction will include
multiple players from diverse sectors, and, to our eyes, it can provide another per-
spective on how an understanding of intangible dynamics can inform strategic
decision-making.

Healthcare
Previous work drew on the metrics of the entire healthcare sector. As just noted,
a report from Bain (Eliades et al., 2012) already defined sectors that might be
included in a healthcare profit pool. And an earlier study (Erickson and Rothberg,
2013a) already gathered and presented some of the intangibles metrics we have
been discussing. These are shown in Table 11.2, along with updated knowledge
data from a new database.
 Intangible Dynamics ◾ 341

Table 11.2 Intangible Dynamics from Healthcare Sectors

Cap/Book Cap/Assets Data
Intelligence Storage/Firm
Sector (SIC #) 10–14 05–09 10–14 05–09 Index (terabytes)

Pharmaceuticals 4.49 4.39 2.06 1.94 64 967 (discrete

(2834) mftg)
831 (process
Diagnostic/ 7.88 4.37 2.73 2.41 14 mftg)
Biological
(2835/6)

Surgical, 3.09 3.43 1.73 2.02 19

medical, dental
instruments
(384)

Lab, optical, 3.90 3.54 1.79 1.52 7

measuring
instruments
(3826/9)

Drugs, 3.78 3.44 1.09 1.16 6 536

wholesale (512)

Medical, dental, 2.83 3.86 1.44 1.43 2

hospital equip/
supplies (5047)

Hospitals (806) 1.79 3.23 0.53 0.61 3 370

Offices, clinics, 1.60 1.79 0.72 0.66 1

various health
(801/5)

Drugs, retail 6.94 2.38 1.04 1.23 1 697

(5912)

Accident/health 1.22 2.29 0.34 0.80 36 870

insurance (632)

Global means 3.59 2.68 1.06 1.02

Source: Updated from Erickson, G.S., and Rothberg, H.N., J. Inform. Knowl. Manag.,
12, 1350031, 2013a.
342 ◾ Analytics and Knowledge Management

One of the interesting things about healthcare is representation from just about
all the different combinations of intangible assets and the domains they represent,
starkly showing how this methodology is capable of distinguishing differences
between the sectors. Further, because the sectors are so different in what they do,
the activities behind the metrics and the competitive considerations they pose are
fairly clear, even at the very general levels we’ll be discussing in this chapter. Finally,
insurance has such unique readings compared to other sectors that its place in the
industry framework is also a straightforward matter to discuss.
As noted in the table, several industry sectors have relatively high levels of all
the intangibles. These include pharmaceuticals, instruments, and diagnostics. They
have highly complex, even chaotic environments demanding accumulation of big
data on research processes, including clinical trials, on highly scrutinized opera-
tional processes, and on marketing relationships (both complicated distribution
chains, third-party payors, and end consumers). Their processes and marketing
need to be highly efficient and often meet rigorous quality standards explaining the
presence of substantial explicit knowledge as well as some tacit knowledge related
to more dramatic improvements. And they still require a steady stream of creative
insights, especially in terms of new products but also including new strategic direc-
tions concerning targeted segments (organizational and consumer), distribution,
marketing, and responses to competitor initiatives. Consequently, ample evidence
exists of tacit knowledge and intelligence.
Alternatively, some sectors show very little of anything at all. Hospitals and
other providers have low levels of big data holdings (though these will likely rise as
the mandate to digitize records takes hold) and there is little evidence of their doing
anything with them. Tobin’s q levels are well under average, so not much is going
on with knowledge development, particularly explicit, and intelligence activity
appears to be virtually nonexistent as well. As pressure builds for U.S. providers to
become more efficient, the data may feed into improvements in explicit knowledge
levels and efficiency, but that is also hard to do when operations are not necessarily
repeatable or systematized (each patient may have their own unique situation and
personal attention is hard to develop knowledge around). Tacit knowledge is likely
present with so many highly skilled workers but, again, it may be hard to share or
to exploit. At least to any degree that would show up in the metrics.
Some sectors show considerable explicit knowledge (and Big Data) but no real
evidence of intelligence activity. As discussed, that likely also means limited tacit
knowledge in spite of the high knowledge metric. The prominent example of this
is the retailers. Along with wholesalers, they generate considerable amounts of data
from their supply chains, their in-store transactions, and their marketing relation-
ships, especially those through loyalty programs and close connections to consum-
ers. Further, those that have added pharmacy benefits management (PBM) to their
activities are developing considerable databases that can draw consumers even
closer. Much of their attention is on continuous improvements in efficiency, run-
ning ever tighter supply chains and retail operations and ever better relationships
 Intangible Dynamics ◾ 343

with consumers, physicians, and insurers paying the bills (explicit knowledge).
While everyone likes a new, insightful idea, their businesses aren’t predicated on
creativity or the types of big steps forward promised by intelligence or even tacit
knowledge.
And, finally, we come to the insurers. Their metrics are indicative of sectors
with a lot of data, evidence of considerable intelligence work, but very low knowl-
edge metrics. This is one of those applications where it is useful to apply both ver-
sions of Tobin’s q as the financial sector has considerable tangible assets, far beyond
what might be seen in other sectors, and so might show artificial differences in the
metric not reflected in actual practice. That would be true of the cap/asset metric
but the cap/book also includes the liabilities associated with the massive tangible
assets, much of which is borrowed in one form or another. So the fact that both
metrics show very low KM activity or success suggests that the sector really is low
when compared to other sectors.
The basic scenario with insurers, and it’s something we’ve seen in just about all
financial services firms, includes massive amounts of data on transactions, individ-
ual client conditions and activities, and marketing relationships. Explicit knowledge
is hard to develop as processes are well-understood and hard to improve while mar-
keting relationships are also well-developed but brand equity is troublesome (very
few consumers are enthusiastic about their insurance companies). Small improve-
ments in process or efficiency aren’t important to the sector. What is sought are big
new ideas, new insights for targeting consumer groups, understanding risk profiles
of different groups, identifying irregularities in the data (fraud, changes in preva-
lent symptoms, changes in prescribing patterns, etc.). Intelligence is pronounced,
including competitive intelligence—when new knowledge or intelligence is so rare,
competitors are going to be interested in uncovering it as quickly as possible.
The core question we’re going to be asking in this section is what opportunities
for growth might be open to insurers, in this case health insurers. The nature of
in-house intangibles indicates considerable holdings of big data of a very specific
nature. There are also indications of experience and competency in managing Big
Data as well as in analyzing it (the analytics and intelligence ratings). The insurance
firms show no particular holdings of or competency in knowledge assets, either
explicit or tacit. Consequently, there would likely be limited potential in venturing
into sectors requiring explicit knowledge or the related efficiencies regarding opera-
tions, supply chains, or external relationships (supply chains, providers, retailers, or
consumers). In this case, insurers would not have good prospects in taking on such
sectors themselves, whether in entry into new sectors, acquisition, or other partner-
ships based on such success requirements.
What healthcare insurers may have to offer are the skills developed in not only
managing Big Data but also in analyzing it for deeper insights (intelligence). This
could be of considerable value to many of the players in sectors with which the insur-
ers interact, including the manufacturers (pharmaceuticals, medical devices) and
the retailers/PBM. Although some data would be duplicative, the insurers have a
344 ◾ Analytics and Knowledge Management

different focus on their data and on how they analyze it, potentially adding value as
a collaborator. Further, the insurers may be of particular help in increasing the capa-
bilities of partners in sectors not so good at managing the intangibles, particularly
Big Data. Although there are obvious relationship issues to overcome with hospitals
and with other providers (as insurers are the payors in most cases), sharing data,
data analysis capabilities and techniques, and acquired insights could benefit all
and seems likely to be one of the strategic avenues with the greatest potential payoff.

Financial Services
The financial services profit pool looks much different from that in healthcare.
There is no dedicated supply chain or manufacturing, nor is there much in the
way of distribution or marketing channels not taken of by the providers them-
selves. Consequently, the profit pool described by Bain consultants (Schwedel and
Rodrigues, 2009) is much more horizontal, encompassing banking, investment
houses, and insurers but little else. This description still includes a number of dif-
ferent industry sectors, as illustrated in Table 11.3 but not much of an industry
value chain. A few distinctions could be made (e.g., insurance providers versus
insurance brokers) but the effective point is made of how different this profit pool
looks compared to the healthcare industry and how that impacts strategic choices
facing participants, especially once one looks more deeply at the intangible assets.
As noted, the results show a very different pattern from what we saw in health-
care industry sectors. The two highlighted sectors are outliers but also somewhat
different in what services they provide. The rest are all mainstream financial ser-
vices, either providing retail banking (commercial banks and savings institutions),
investment services (security brokers and Real Estate Investment Trusts (REITs)
REITs), or insurance (life, health, property). All of these involve very large data-
bases. These databases include customer knowledge such as personal characteristics,
transaction records, and relationship tracking (including all communications). All
also track data on economic indicators, market financial results, and other world,
country, or industry-wide information. In the McKinsey report, financial services
take up most of the top spots in big data holdings and by a wide margin.
The knowledge data are a different story. Again ignoring the shaded rows for now,
the rest of the sectors are uniformly well under the average cap/book and cap/assets
ratios seen across the databases. This is true when looking at both metrics. Here, cap/
assets may be artificially low just because of the huge level of tangible financial assets
held by these types of institutions—inflating the denominator of the ratio. But the
same pattern is also seen in the cap/book ratio which would level out the bias by tak-
ing into account any financial assets borrowed or employed on behalf of a customer
who actually owns them. If there is some artificial deflation of the results, it is mini-
mal and certainly doesn’t change the overall conclusions. In financial services sectors,
there is very little evidence of knowledge, particularly explicit knowledge.
 Intangible Dynamics ◾ 345

Table 11.3 Intangible Dynamics for Financial Services

Cap/Book Cap/Assets Data
Storage/
Intelligence Firm
Sector (SIC #) 10–14 05–09 10–14 05–09 Index (terabytes)

Commercial 1.13 1.61 0.12 0.14 45 1931

banks (6020)

Savings 1.16 1.14 0.16 0.11 18

institutions
(6035/6)

Security 1.63 1.52 0.36 0.19 13 3866

brokers and
dealers (6211)

Investment 2.54 4.73 1.18 1.57 3

advice (6282)

Life insurance 0.80 1.12 0.10 0.11 22 870

(6311)

Accident/ 1.22 1.49 0.34 0.48 54

health,
hospital and,
medical
services
(6321/24)

Fire, marine 1.00 1.22 0.28 0.34 29

and casualty
(6331)

Insurance 1.91 3.94 1.28 1.21 4

agents and
brokers
(6411)

Real estate 2.36 1.45 0.91 0.47 16 3866

investment
trusts (6798)

Global means 3.59 2.68 1.06 1.02

346 ◾ Analytics and Knowledge Management

There is, however, considerable evidence of intelligence. Given the low knowl-
edge metric, one would conclude that much of this rating is attributable to insight
or analytical abilities present in these sectors but not tacit knowledge. Assuredly,
new tacit knowledge or intelligence insights are rare, valuable, and come by means
of a process difficult or impossible to teach others. Organizations may set up struc-
tures to feed intangible inputs, especially data, to analysts and encourage insight
and creativity, but success depends on hiring and retaining the right people more
than on how they are taught their responsibilities.
This squares with what we know about these industry sectors. As noted, they
manage huge amounts of tangible financial assets. Doing so requires little knowl-
edge about efficiency of supply chains, operations, or customer relationships
(indeed, consumers often actively dislike their banks and insurance companies,
so brand equity is minimal). They do know how to execute transactions. They do
know how to report to customers, regulators, and shareholders. But much of that is
established and well-known, there is little new to discover in the nature of explicit
(or tacit) knowledge improving activities or relationships.
What can be new are creative insights, crafting new investment strategies or
new insurance risk profiles and products. When new knowledge and insights are
so rare, they take on additional value, so the rewards for successful analytics are
considerable. This is seen in the level of intelligence metric. A new, unique strategy
is highly valued so other firms are extremely interested in uncovering competitor
discoveries as quickly as possible. The competitive intelligence levels are quite high,
reflecting this reality as well as the considerable weight all of these financial services
sectors place on new solutions discovered through an effective analytics or intel-
ligence process.
The exceptions validate the rule. The shaded rows have both higher knowledge
and lower intelligence metrics. Both of these sectors, investment advisor and insur-
ance brokers, depend more on individuals with knowledge that can be applied to
variable conditions (advising consumers according to their specific situations and
needs). For insurance brokers, this is fairly obvious as agents actually selling the
insurance tend to be smaller concerns with close, personal client relationships and
a need to match the client with the right insurance. They learn what works for
what clients in what circumstances may have access to Big Data but may not know
what to do with it, and rely more on their own assessment and conclusions than on
repeatable patterns of action.
The investment advisors are more complex and, in some ways, more interest-
ing. The firms making up this sector are quite a mixed bag. They range from
high-powered investment firms (Blackstone, Carlyle, Invesco) to retail brokerages
(Charles Schwab, T. Rowe Price). Generally, the former group has low knowledge
metrics, similar to those seen with security brokers. The emphasis is again on Big
Data and unique insights more than repeatable explicit or tacit knowledge. The
latter group has the higher knowledge metrics as they do provide trade executions
to retail buyers, a repeatable pattern that can yield knowledge based on segmented
 Intangible Dynamics ◾ 347

groups of customers. These firms also depend on close relationships with clients,
something that we know requires good levels of customer knowledge and is again
usually apparent in high explicit knowledge results.
To return to the overall point of this section, what do these results tell us about
the strategic potential of insurers in this industry? Unlike the healthcare industries,
in which insurers had little in common with any of the other industry sectors, the
results are quite similar here across the categories. One could see substantial move-
ment across sectors here as the different players have similar intangibles competen-
cies. All have some substantial level of Big Data and apparently know what to do
with it. All also have some substantial level of analytical or intelligence ability and,
again, apparently know what to do with it. While partnerships to share data and
insights are possible, as in healthcare, vertical movements across sectors in a more
aggressive manner are also possible.
One indicator of this potential comes from recent news reports about insur-
ance firms moving into pension management. Based on the intangibles, this makes
a great deal of sense. As repeatedly noted, Big Data is present and important
throughout all these sectors, so insurers’ capabilities in managing large databases
concerning client descriptions, activities, and more macro trends would be applica-
ble across sectors. Perhaps even more importantly, insurers deal routinely with both
individual retirement plans and actuarial data. Their experience with such data and
their ability to find new insights in it after conducting deep analysis would be well-
suited to pension management. Most also have experience with finding suitable
investment opportunities for held funds as well as efficiency in client relationships
(even if not overly warm and fuzzy). The intangibles results show a strong fit for
this sort of move.

Automated Driving
Although a profit pool study for this burgeoning industry hasn’t yet been compiled,
work has been done on active and announced participants so we do have some
sense of who might be involved. Further, Navigant Research (Abuelsamid et al.,
2017) has released a report assessing the prospects of known aspirants according to
a checklist of success criteria. Other observers might adjust the criteria or provide
different ratings for individual firms, but the structure is there for us to consider the
field with respect to intangibles.
The Navigant report, based on public announcements of the represented firms,
includes the industry sectors listed in Table 11.4. To these, we have added both
Apple (no announcement but substantial evidence of intentions) and Intel, which
recently announced its own interest and a partnership with Mobileye. Apple’s finan-
cial filings listed it as Computers (SIC 357) during the earlier reporting period but
switched to Phones (SIC 3663) during the latter. The company, of course, competes
in both sectors though its emphasis and revenue and profit streams have changed.
Both sectors are listed here.
348 ◾ Analytics and Knowledge Management

Table 11.4 Intangible Dynamics for Automated Driving Sectors

Cap/Book Cap/Assets Data
Storage/
Industry Sector Intelligence Firm
(SIC) 10–14 05–09 10–14 05–09 Index (terabytes)

Auto 8.45 0.80 0.82 0.47 9 967

manufacturing
(3711)

Auto 11.40 2.10 1.04 0.39 10

components
(Delphi, ZF)
(3714)

Computers 5.63 4.48 0.74 1.58 22

(Apple) (357)

Phones (Apple) 2.57 2.71 1.18 1.56 17

(3663)

Semiconductors 4.00 4.80 1.66 1.97 11

(Intel) (3674)

Software/Web 5.02 3.48 2.30 2.08 19 1792

(Alphabet/
Waymo, Baidu,
nuTonomy)
(7370)

Auto rental 3.28 1.64 0.43 0.30 3 801

(Uber) (7510)

Fire, marine, 1.22 1.00 0.34 0.28 29 870

casualty
insurance (6331)

Database average 3.59 2.68 1.06 1.02

For the record, the Navigant “leaders,” those best placed to compete in the
new area, are mainly traditional auto manufacturers (Ford, GM, Toyota, BMW,
Tesla, etc.). The reasons behind their conclusions are probably best seen in the
criteria used for assessment, including both strategy variables (vision, go-to-
market strategy, partners, production strategy, and technology) and execution
variables (sales, marketing, and distribution; product capability; product quality
and reliability; product portfolio; and staying power). The details don’t match
up exactly, but one could easily associate much of the strategy component with
 Intangible Dynamics ◾ 349

tacit knowledge and intelligence. Alternatively, execution will often have more
to do with explicit (and some tacit) knowledge.
What does the data tell us? The automakers have relatively low knowledge lev-
els across the board, except for the very high 2010–2014 cap/book ratio. In looking
into the data, this result is chiefly due to the presence of Tesla and is not mirrored
in the cap/assets ratio. This is a case where using both metrics is useful. Tesla is
highly leveraged, so the book value in the first calculation is artificially low. Cap/
asset is probably the better and seemingly more consistent metric to pay attention
to in this case though Tesla’s considerable store of intangible assets should also be
kept in mind. The auto industry has improved markedly from the earlier to the
later time period in the knowledge metrics, recall that the earlier period included
the 2008 financial downturn. The intelligence variable is average (anything in
double digits is getting into more aggressive intelligence activity). Big Data is
present, in substantial quantities but at a level that is also about average across
all industries. All in all, the auto manufacturers have some capability in explicit
knowledge (improving but still below average), evidence of some tacit knowledge
and intelligence, and some Big Data. But nothing is outstanding, especially rela-
tive to the other industries. The manufacturers should be competent at supply
chains, manufacturing, distribution, and customer relationships. They should also
have some abilities in more creative insights such as new product development.
But there is also nothing here that would scare anyone else off, perhaps why the
field is so full of potential competitors from different sectors and interesting com-
binations of old-line manufacturers and new-line software and other firms.
Tesla is a bit of a special case. As this was being written, the electric vehicle
manufacturer’s market cap passed GM’s for the first time, making it the most valu-
able U.S. auto company. As noted, the firm’s knowledge metric is considerably
higher than other competitors, particularly in the cap/book version but also in
cap/asset. Even though we can’t measure it directly, the firm likely also has higher
tacit knowledge and, perhaps, intelligence levels given its formidable research and
development efforts and presence of key players such as the founder. Moreover, the
firm has specialist knowledge in certain areas (batteries) that may or may not be a
feature of automated driving. This methodology gives a nice snapshot of the full
industry sector but does require a deeper dive into the individual characteristics of
firms to fully understand what might be happening, and Tesla might be explored
in more detail.
The next four sectors, highlighted in the table, are a mix of manufacturing and
services (software). All have relatively high knowledge metrics though the phone
sector is showing signs of decline. Knowledge in the efficiency of design, execution,
and delivery of computers, phones, semiconductors, and software is important in
all these fields, even when some parts of that chain are outsourced (e.g., Apple). A
number of high-powered brands are also present in these sectors (Apple, Google,
Intel, Microsoft), and so relational capital is also good and likely pushes up the
knowledge metric. Intelligence scores are also high in all the sectors and big data
350 ◾ Analytics and Knowledge Management

is present. Essentially, firms in these sectors possess relatively high levels of all the
intangible assets, potentially adaptable to different circumstances and making dan-
gerous competitors of key players in each.
Uber, slotted in the auto rental sector, is another special case. To some extent,
it is a special player in the rental sector with a different approach than traditional
agencies (who usually rent the entire car rather than just the single ride). Here, its
financials are not included in the results of the full sector as it was not publicly
traded when the data was reported. On the other hand, most of the major players
in the sector are moving to new service models, whether car sharing, ride sharing,
shorter-term rentals (e.g., hourly), or some other variation. So Uber (and Lyft’s)
impact on the sector should start to be visible.
What the data in auto rentals show are low but increasing knowledge levels.
Once again, given the large fleet of tangible assets that may be debt-financed, the
cap/asset ratio may be the more accurate metric though both cap/asset and cap/
book generally agree. Intelligence activity is very limited, and big data is not overly
pronounced either. One area where Uber may have an impact is in Big Data as its
network thrives on collecting rider and driver data, creating greater efficiency by
matching supply and demand. And this may drive up the knowledge metrics over
time as managers get a better handle on logistics and competitors are forced to catch
up. But, overall, the intangibles capabilities of this sector just don’t suggest a serious
player. Ride-sharing firms may have interest in autonomous vehicles but seem more
likely to be customers or partners for the technology rather than developers.
Which brings us to the insurance sector once again. Here we have included the
Fire, Marine, Casualty sector that would include auto insurers but, again, recall
that just about all insurers have similar intangible profiles. Once again, lots of data.
Huge amounts of data compared to some other prospective participants in the
field. And, again, not much in the way of explicit or tacit knowledge, according to
the knowledge metric. But what is there is some intelligence ability combined with
the Big Data, suggesting a competency in accumulating, processing, and finding
insights in databases.
There is no indication of the interest of any insurance firm in direct participa-
tion in the autonomous driving field. That might be a good thing. There is also
no indication in the metrics of any capability to participate in development of key
pieces of the technology such as sensors, software, vehicle manufacturing, or so
forth. The car insurance firms, however, do have a vested interest in the progress of
the technology, especially in terms of how it impacts accident rates, their severity,
and where liability might be placed. As such, the auto insurance firms might be
very valuable collaborators in this field, contributing their big databases on driv-
ers, driving activity, and, particularly, accidents. The software and artificial intel-
ligence systems that will guide autonomous driving cars are being developed in a
trial-and-error manner by putting the vehicles on the road and teaching them to
recognize circumstances and make the right decisions. While this makes sense, the
process could likely be enhanced by incorporating the data the insurers already
 Intangible Dynamics ◾ 351

have on accident circumstances and correct or incorrect choices made by drivers.

The insurers have potentially valuable data. They also have the analytical abilities
to properly study the data and find the insights that more direct participants in the
autonomous driving field could find useful. Considerable press has suggested that
safe, mistake-free, driverless cars could kill the insurance sector. But since their
intangibles are something not necessarily matched in-house by many of the would-be
competitors, the insurers may instead take a role in the development process and
find a way to be useful partners heading into the future.

Conclusions
Even though relatively young disciplines, KM, and IC have already faced a couple of
waves of growth and decline in interest over the past few decades. The advent of inter-
est in Big Data and business analytics and intelligence presents another possible threat
or opportunity for the field. In some ways, the burgeoning investment in big data
systems, including the cloud, and associated business analytics efforts have left KM
installations behind. The possibilities of gathering massive amounts of data and apply-
ing it to make better and more informed decisions at all levels are intriguing. And the
promise may be more compelling than some of the results seen from KM installations.
But opportunities are also present. The central question in KM and IC has
always been about the systems used to identify, measure, and manage intangible
assets. With knowledge, those assets originated in individuals and the systems
had to encourage the human resources of organization to contribute their knowl-
edge while being open to employing that gained from others. The complexities of
humans interacting with IT systems or other tools have been at the core of much of
the discussion over the past 30 years.
From that perspective, both big data systems and business analytics efforts can
be seen as extensions, with different kinds of intangibles. As such, KM learnings
may have something important to contribute, not only in what the systems look
like for managing the intangibles but also how to execute the human element that
makes the systems actually work. Further, KM and IC have been employed to help
understand competitive environments, how intangibles are assessed and exploited
in different circumstances in different industries. These sorts of applications can
also be extended to the wider range of intangibles, not just knowledge varieties but
also data, or information, and intelligence.
This chapter has brought together work from a variety of disciplines to explore
this strategic perspective in more detail. In particular, we have presented an updated
conceptualization of the intangibles hierarchy, a step that helps to explain the simi-
larities and differences between data and information, explicit knowledge, tacit
knowledge, and insight and intelligence. Traditional KM and IC covers explicit and
tacit knowledge but the extensions to data and information, with connections to
Big Data, and to insight and intelligence with its connections to business analytics,
352 ◾ Analytics and Knowledge Management

are useful in framing the full picture of contemporary intangible asset assessment
and usage. This perspective adds even more complexity to our understanding of
competitive environments and what intangibles are most effective in them but it
also brings us closer to the reality of today’s business battlegrounds.
With a firm grasp of the full range of intangibles, we can also look to accurately
measure them. In this chapter, we have demonstrated how to do so in specific
industry sectors and, in some cases, by individual firms. One can get a sense of the
competitive capabilities of those individual firms but here we have looked primarily
at what it takes in terms of intangible assets to be effective in the different sectors.
From a strategic planning perspective, this approach provides guidance as to what
a firm needs in its own sectors as well as what it might add in others, including
whether it possesses the intangibles levels and management skills to be competitive
on its own in a new environment.
With that in mind, we have provided analysis of three industries, guided by
profit pool structures identifying the key sectors in two of those industries. Given
the presence or potential presence of a single industry sector, insurance, we can
more easily see how the methodology provides guidance to decision makers. In
healthcare and automated driving, the insurers do not have the intangibles to be
competitive players but might have a role to play in providing Big Data and busi-
ness analytics capabilities to other industry challengers from other sectors (who
may not have such intangibles or skill in managing them). In financial services, the
intangibles capabilities of all types of insurers look similar and are also very much
like those in other sectors (banking, investment services). In such a case, insurers
may have the potential to be serious competitors themselves in tangential sectors.
Such analysis, in a short chapter like this, is necessarily only at the surface.
But the approach illustrates what decision makers in these industry sectors, with
considerably more knowledge of the competitive details behind the metrics, could
do. The intangibles metrics provide guidance as to the state of affairs in the sector
and alerts such decision makers where to focus their efforts. Their own, more spe-
cific knowledge can then lead to the deeper insights that help make better strategic
choices.

References
Abuelsamid, S., Alexander, D., and Jerram, L. (2017). Navigant research leadership report:
Automated driving (executive summary). Available at https://www.navigantresearch.
com/wp-assets/brochures/LB-AV-17-Executive-Summary.pdf.
Ackoff, R. (1989). From data to wisdom, Journal of Applied Systems Analysis, 16, 3–9.
Argyris, C. (1977). Double loop learning in organizations, Harvard Business Review,
September/October, pp. 115–125.
Argyris, C. (1992). On Organizational Learning, Blackwell, Cambridge, MA.
Argyris, C. (1993). Knowledge for Action, Jossey-Bass, San Francisco, CA.
 Intangible Dynamics ◾ 353

Bontis, N. (1999). Managing organizational knowledge by diagnosing intellectual capi-

tal: Framing and advancing the state of the field, International Journal of Technology
Management, 18(5–8), 433–462.
Brown, J.S., and Duguid, P. (1991). Organizational learning and communities-of-practice:
Toward a unified view of working, learning, and innovation, Organizational Science,
2(1), 40–57.
Brown, L., and Boudes, T. (2005). The use of narrative to understand and respond to com-
plexity, E:CO, 7(3–4), 32–39.
Chen, H., Chiang, R.H., and Storey, V.C. (2012). Business intelligence and analytics: From
big data to big impact, MIS Quarterly, 36(4), 1165–1188.
Crocker, A. (2014). Combining strategic human capital resources: The case of star knowl-
edge workers and firm specificity, Academy of Management Proceedings, January,
p. 10705.
Edvinsson, L., and Malone, M. (1997). Intellectual Capital, Harper Business, New York.
Eliades, G., Retterath, M., Hueltenschmidt, N., and Singh, K. (2012). Healthcare 2020.
Available at http://www.bain.com/Images/BAIN_BRIEF_Healthcare_2020.pdf.
Erickson, G.S., and Rothberg, H.N. (2012). Intelligence in Action: Strategically Managing
Knowledge Assets, Palgrave Macmillan, London, UK.
Erickson, G.S., and Rothberg, H.N. (2013a). Alternative metrics for assessing knowledge
assets, Journal of Information & Knowledge Management, 12(4), 1350031.
Erickson, G.S., and Rothberg, H.N. (2013b). A strategic approach to knowledge develop-
ment and protection, Service Industries Journal, 33(13/14), 1402–1416.
Erickson, G.S., and Rothberg, H.N. (2016). Intangible dynamics in financial services,
Journal of Service Theory & Practice, 26(5), 642–656.
Firer, S., and Williams, S.M. (2003). Intellectual capital and traditional measures of corpo-
rate performance, Journal of Intellectual Capital, 4(3), 348–360.
Grant, R.M. (1996). Toward a knowledge-based theory of the firm, Strategic Management
Journal, 17(Winter), 109–122.
Groysberg, B., Lee, L.-E., and Abrahams, R. (2010). What it takes to make “star” hires pay
off, Sloan Management Review, 51(2), 57–61.
Kogut, B., and Zander, U. (1992). Knowledge of the firm, combinative capabilities, and the
replication of technology, Organization Science, 3(3), 383–397.
Kurtz, C.F. (2009). The wisdom of clouds, white paper. Available at http://www.cfkurtz.
Kurtz, C.F., and Snowden, D.J. (2003). The new dynamics of strategy: Sensemaking in a
complex-complicated world, IBM Systems Journal, 42(3), 462–483.
Laney, D. (2001). 3D data management: Controlling data volume, velocity and variety.
Available at November 1, 2013, http://blogs.gartner.com/doug-laney/files/2012/01/
ad949-3D-Data-Management-Controlling-Data-Volume-Velocity-and-Variety.pdf.
Liebowitz, J. (2005). Linking social network analysis with the analytical hierarchy process for
knowledge mapping in organizations, Journal of Knowledge Management, 9(1), 76–86.
Manyika, J., Chui, M., Brown, B., Bughin, J., Dobbs, R., Roxburgh, C., and Hung Byers, A.
(2011). Big Data: The Next Frontier for Innovation, Competition and Productivity,
McKinsey Global Institute, New York.
Marr, B., Schiuma, G., and Neely, A. (2004). The dynamics of value creation: Mapping
your intellectual performance drivers, Journal of Intellectual Capital, 5(2), 312–325.
Matson, E., Patiath, P., and Shavers, T. (2003). Stimulating knowledge sharing:
Strengthening your organizations’ internal knowledge market, Organizational
Dynamics, 32(3), 275–285.
354 ◾ Analytics and Knowledge Management

McAfee, A., and Brynjolfsson, E. (2012). Big data: The management revolution, Harvard
Business Review, 90(10), 60–66.
McEvily, S., and Chakravarthy, B. (2002). The persistence of knowledge-based advantage:
An empirical test for product performance and technological knowledge, Strategic
Management Journal, 23(4), 285–305.
Nahapiet, J., and Ghoshal, S. (1998). Social capital, intellectual capital, and the organiza-
tional advantage, Academy of Management Review, 23(2), 242–266.
Nonaka, I., and Takeuchi, H. (1995). The Knowledge-Creating Company: How Japanese
Companies Create the Dynamics of Innovation, Oxford University Press, New York.
Polanyi, M. (1967). The Tacit Dimension, Doubleday, New York.
Rothberg, H.N., and Erickson, G.S. (2005). From Knowledge to Intelligence: Creating
Competitive Advantage in the Next Economy, Elsevier Butterworth-Heinemann,
Woburn, MA.
Rothberg, H.N., and Erickson, G.S. (2017). Big data systems: Knowledge transfer or intel-
ligence insights, Journal of Knowledge Management, 21(1), 92–112.
Schwedel, A., and Rodrigues, A. (2009). Financial services’ shifting profit pool. Available
at http://www.bain.com/Images/BB_Financial_services_shifting_profit_pools.pdf.
Senge, P. (1990). The Fifth Discipline, Doubleday, New York.
Simard, A. (2014). Analytics in context: Modeling in a regulatory environment, in
Rodriguez, E. and Richards, G. (Eds.), Proceedings of the International Conference
on Analytics Driven Solutions 2014. Harvard University Press, Cambridge, MA,
pp. 82–92.
Stewart, T.A. (1997). Intellectual Capital: The New Wealth of Organizations, Doubleday,
New York.
Sveiby, K.E. (2010). Methods for measuring intangible assets. Available at http://www.
sveiby.com/articles/IntangibleMethods.htm.
Tan, H.P., Plowman, D., and Hancock, P. (2007). Intellectual capital and the financial
returns of companies, Journal of Intellectual Capital, 9(1), 76–95.
Teece, D.J. (1998). Capturing value from knowledge assets: The new economy, markets for
know-how, and intangible assets, California Management Review, 40(3), 55–79.
Thomas, J.C., Kellogg, W.A., and Erickson, T. (2001). The knowledge management puzzle:
Human and social factors in knowledge management, IBM Systems Journal, 40(4),
863–884.
Tobin, J., and Brainard, W. (1977). Asset markets and the cost of capital, in Nelson, R. and
Balassa, B. (Eds.), Economic Progress, Private Values, and Public Policy: Essays in Honor
of William Fellner. North Holland, Amsterdam, the Netherlands, pp. 235–262.
Verdun, J. (2008). The last mile of the market: How network technologies, architectures
of participation and peer production transform the design of work and labour, The
Innovation Journal: The Public Sector Innovation Journal, 13(3). Available at https://
www.innovation.cc/.
Wernerfelt, B. (1984). The resource-based view of the firm, Strategic Management Journal,
5(2), 171–180.
Zack, M.H. (1999). Developing a knowledge strategy, California Management Review,
41(3), 125–145.
Zander, U., and Kogut, B. (1995). Knowledge and the speed of transfer and imitation of
organizational capabilities: An empirical test, Organization Science, 6(1), 76–92.
Chapter 12

Analyzing Data and

Words—Guiding
Principles and
Lessons Learned
Denise A. D. Bedford

Contents
Introduction ......................................................................................................356
Conceptual Framework .....................................................................................357
Research and Business Goals (Why?) .................................................................359
What Are You Trying to Achieve? What Is Your Research or Business Goal?....359
What Good Practices and Good Practice Models Exist? ........................... 360
How Will You Measure the Results of Your Analysis?............................... 360
What Level of Risk Are You Willing to Assume? .......................................361
What Level of Investment Are You Willing to Make?................................361
Is This a Project or Enterprise-Level Goal? ................................................362
Understanding Why in Context—Use Case Scenarios ..............................362
How We Use the Tools—Analysis as a Process ...................................................370
What Analytical Method Is Best Suited to the Goals? ...............................371
Quantitative Analysis and Data Analytics ......................................371
Qualitative Analysis and Language Based Analytics .......................372
Mixed Methods Analysis and Variant Sources................................374
When Is a Quantitative Analysis Approach Warranted? ............................375
When Is a Qualitative Analysis Approach Warranted?...............................375

355
356 ◾ Analytics and Knowledge Management

When Should We Choose a Mixed Methods Approach? ...........................375

Which of These Analytical Methods Are Supported by Tools?...................376
Which of These Analytical Methods Are Not Supported by Tools? ...........376
What Opportunities Are There for Mixing Analytical Methods? ...............377
How in Context—Use Cases ....................................................................377
What Sources Are We Analyzing? ......................................................................381
What Is the Nature of Language, Information, and Knowledge We’re
Using as Source Evidence? ........................................................................382
What Meaning or Understanding Are We Deriving from the Language? ......384
What Linguistic Registers Are Represented in the Source Evidence? .........384
What Knowledge Structures Are Represented in the Source Evidence? .....385
What Types of Semantic Methods Are Available for Us to Work With? ....386
What Does the Analytical Workflow Look Like for Each of These
Methods? .................................................................................................386
Do These Methods Offer the Possibility of Reducing or Eliminating
the Subjective Interpretation Element? .....................................................386
What Is the Return on Investment for the Analysis? .................................387
What Level of Effort Is Involved in Doing a Rigorous Analysis? ...............387
What Types of Competencies Are Needed to Support the Analysis? ..........387
What Is the Feasibility of the Analysis Without Semantic Methods? .........387
Understanding Sources in Context—Use Case Examples..........................388
Fitting Tools to Methods and Sources ................................................................396
Reviewing Your Tool Choices ...................................................................397
Reviewing Your Successes .........................................................................397
Exposing Your Failures..............................................................................397
Lessons Learned and Future Work .....................................................................398
Conclusions ..................................................................................................... 400
References .........................................................................................................401

Introduction
Business analytics and text analytics are not new. Work on the core analytics capa-
bilities dates back to the 1950s and 1960s. What has changed in 2017 is the com-
puting capacity we have—in the business and the research environment—to apply
analytics. The increased capacity has created an expanded set of expectations for
what is possible. It has also created opportunities to explore business and research
questions we previously thought impossible due to scale, scope, cost, or reliability.
The expanded capacity has allowed us to explore the ways in which humans lever-
age language to create and represent knowledge. Knowledge management meth-
ods and core concepts are critical to the intelligent use of analytics. The expanded
foundation makes it possible to do much more than detect clusters and patterns in
text. This new capacity holds great promise for advancing the discipline of knowl-
edge management. Knowledge management has traditionally relied on qualitative
 Analyzing Data and Words—Guiding Principles ◾ 357

methods that rely on human manual interpretation and analysis. The discipline
of knowledge management, which has traditionally relied on qualitative methods,
can be enhanced and expanded through the intelligent use of analytics. These
expectations and opportunities are achievable, but only if we approach analytics
intelligently. New linguistic and knowledge-based tools allow us to use machines
to begin to truly understand text. Understanding, though, requires thoughtful
design, investments, error and risk management, a fundamental understanding
of qualitative methods and their machine-based transformation, a fundamental
understanding of language and linguistics, and a willingness to navigate today’s
volatile “analytics” market. This chapter considers how we can leverage both of
these disciplines to meet expectations and new demands. This chapter offers guid-
ance for navigating these issues in the form of key questions and lessons learned.

Conceptual Framework
Over the past 35 years, four focus points have helped the author navigate the choice
and use of analytical methods and tools to a range of projects. The focus points
include: intent and focus of the project (why); the nature of analysis (how); sources
we analyze and use for analysis (what); and tools and methods that support the
kind of analysis we’re doing. The remaining sections of the chapter walk through
the key questions for each dimension. Each focal point is supported by a set of criti-
cal thinking questions (Table 12.1). The 27 questions are presented in sequential
order. Lessons learned suggest that addressing each of these questions individually
increases the probability of a successful effort.
In the sections that follow, we explore each of these key questions in the context
of seven real world use cases. These use cases have all ended in success but could
have resulted in significant failures had we not considered the key questions. The
use cases are identified below.

◾ Use Case 1: Causal Relationships between Arab Spring and Release of Wiki-
leaked Cables (manuscript in process)
◾ Use Case 2: Emotional and Social Tone of the Discourse of the 2012 U.S.
Presidential Campaign (Bedford, 2012b)
◾ Use Case 3: Language and Knowledge Structures of Mathematical Learning
(Bedford & Platt, 2013)
◾ Use Case 4: Knowledge Transfer Practices among Beekeepers (Bedford &
Neville, 2016)
◾ Use Case 5: Precision of Automated Geographical Categorization (Bedford, 2012)
◾ Use Case 6: Analysis of Physician-Patient Communication (Bedford, Turner,
Norton, Sabatiuk, & Nassery, 2017)
◾ Use Case 7: Analysis and Detection of Trafficking in Persons (Bedford,
Bekbalaeva, & Ballard, 2017)
358 ◾ Analytics and Knowledge Management

Table 12.1 Key Questions by Dimensions of the Framework

Dimension Key Questions

Research and 1. What are you trying to achieve?

business 2. What good practice and theoretical models exist?
goals—why 3. How will you measure the results?
4. What level of risk are you willing to assume?
5. What level of investment are you willing to make?
6. Is this a project or enterprise level of effort?

Analytical 7. What analytical method is best suited to the goal?

methods—how 8. When is a quantitative analysis approach warranted?
9. When is a qualitative analysis approach warranted?
10. When should we choose a mixed methods approach?
11. Which of these methods are supported by tools?
12. Which of these methods are not supported by tools?
13. What opportunities are there for mixing tools?

Sources of 14. What is the nature of language, information, and

evidence knowledge source evidence?
15. What meaning or understanding are we deriving from
the language?
16. What linguistic registers are represented in the source
evidence?
17. What knowledge structures are represented?
18. What types of semantic methods are available for us to
work with?
19. What does workflow look like for each of these
methods?
20. Is there a possibility of reducing or eliminating the
subjective interpretation element?
21. What is the return on investment for the analysis?
22. What level of effort is involved in doing a rigorous
analysis?
23. What types of competencies are needed to support
the analysis?
24. What is the feasibility of the analysis without semantic
methods?

Tools 25. Reviewing the tools used

26. Reviewing your successes
27. Reviewing your failures
 Analyzing Data and Words—Guiding Principles ◾ 359

Research and Business Goals (Why?)

Every organization will use analysis. We use analytics for many traditional reasons
including making better decisions, achieving better research results, discovering
ways in which our organizations can be more efficient or effective. We also use
analytics because they allow us to leverage machines to achieve what is beyond the
scale and scope of human manual analysis, and in some cases to remove the inherent
subjective interpretation humans bring to any analysis. Twentieth century indus-
trial economy organizations relied primarily on traditional analyses— quantitative
methods and quantitative sources. Machine-based analytics offer the potential to
go beyond what we have traditionally expected. An important new direction is that
offered by knowledge management. Twenty-first century knowledge economy orga-
nizations require new and more robust analytical methods. Knowledge management
uses all of the traditional methods and sources to “…. make the organization act as
intelligently as possible and realize the best value from its knowledge assets, that is,
to create a learning organization that is capable of measuring, storing, and capital-
izing on the expertise of employees to create an organization that is more than the
sum of its parts” (Bollinger & Smith, 2001).
The need for analysis to support knowledge management goes beyond what has
traditionally been available. There are significant opportunities for leveraging busi-
ness analytics to create a knowledge organization and for using knowledge man-
agement methods to enrich business analytics. Organizations need to identify and
leverage their business-critical knowledge. Business capability and modeling meth-
ods are qualitative methods that leverage qualitative sources. Quantitative methods
may provide partial support, but additional qualitative machine-based methods
are needed. Organizations have vast stores of qualitative sources (e.g., text, speech,
audio in a variety of languages) which hold important knowledge and insights that
can help us to achieve Bollinger and Smith’s goals. To derive meaning from these
sources, we must be having a working understanding of language.
Whether you are a forward-looking or a traditional organization, your use of
analytics should be guided by what you are trying to achieve, what good models or
practices you have to work with, the level of risk you are willing to accept, the level
of investment you are willing to make, and the scale and scope of the application.

What Are You Trying to Achieve? What Is Your

Research or Business Goal?
To prosper, an organization must leverage all of the resources available to it. In an
increasingly competitive world, all organizations—universities, research institutes,
global private companies, non-governmental organizations—understand that it is
no longer acceptable for individuals to make business decisions or define research
projects based on “gut feelings,” “knee jerk reactions,” or personal experience.
360 ◾ Analytics and Knowledge Management

Business and research goals define the purpose and reason a company, organization
or institution exists.

What Good Practices and Good Practice Models Exist?

When a new tool or method is developed, we try to fit it into the existing reper-
toire. More often than not a new method or tool complements or supplements what
exists. It is rarely the case—except in fundamental theoretical research—that an
entire way of thinking or seeing is replaced. By definition, complementary and
supplementary methods and processes add to or support what exists. Where the
focus is on new machine-based methods, in all probability we have good human-
based practical and theoretical models. The new process or method gives us the
opportunity to test what exists on a broader scale, to expand the scope of analysis,
to achieve a comparable result in a much more cost-effective or cost-efficient man-
ner. The availability of new tools does not replace or eliminate existing models and
methods. In all cases, the tool is the last decision or choice we make. We follow
basic research and business practices to define our goals, lay out our questions, and
consider what knowledge already exists. We need to build on, enable, or enhance
existing knowledge wherever it exists.
The important lesson learned here is that an effective model applied manually
can and should serve as the basis for a machine-based model. If there is a well-
formulated theoretical model, the first question should be how that model can
be adapted to machine processing. This can often be a critical success factor. In
fact, in some cases, the machine transformation of a model that has been applied
manually for years can provide a rigorous test of that model. It does not make good
business or research sense to disregard a good theoretical model because it is not in
machine-processable form. Just because you have a new technology or tool to work
with does not mean you can disregard well-formulated and long-standing business
rules, theory, or reference models.

How Will You Measure the Results of Your Analysis?

Thinking about measurement forces us to transform our thoughts and ideas into
quantitative results and to express our observations numerically. Measurement
applies to both quantitative and qualitative analyses. The goal of measurement
should be the reduction of uncertainty and bias. Measurement of quantitative is
supported by well-formed and time-tested statistical and mathematical methods.
Measurement of qualitative sources is more challenging because it has tradition-
ally involved human review and interpretation of results. Even the most seasoned
researcher or decision maker finds it challenging to hold back their experiences,
beliefs, and perspectives. The new tools do not preempt or negate the need to mea-
sure results, to define potential errors, or to dismiss concerns of validity and reli-
ability. Rather, new tools will require new measures. Measurement is a core element
 Analyzing Data and Words—Guiding Principles ◾ 361

of ensuring that we achieve our business and research goals. Errors occur in degrees.
Defining our tolerance for errors is an import part of measurement. Two consider-
ations that help us to manage errors are reliability and validity. Both reliability and
validity errors are possible with the new analytical tools. In fact, both may increase
if we chose tools and sources incorrectly.
Reliability speaks to the consistency in measure and considers whether you
would obtain the same result if you repeated the analysis with all factors remaining
constant (Babbie, 1989; Carmines & Zeller, 1979; Kirk & Miller, 1986). A simple
example of a reliability error occurs when we apply quantitative analytical tools
to dynamic text or language. The reliability and generalizability of results is com-
promised on each time the corpus of text changes. Validity is understood as truth
in measurement and tells us whether we measured what we intended to measure
(Adcock, 2001; Guion, 1980; King, Keohane, & Verba, 1994). Validity is easier to
measure in quantitative analyses than it is in qualitative analyses (Atkinson, 2005;
Reason and Bradbury, 2001; Eriksson & Kovalainen, 2015; Seale, 1999). A simple
example of a validity error occurs when we apply quantitative analytical tools to
text or language without first developing a deep understanding of the linguistics
or the knowledge structures inherent to the text. The simplest examples of errors,
though, are type 1 and type 2 errors.

What Level of Risk Are You Willing to Assume?

Analytical tools are not free of risk. It is important to understand the nature and
extent of risk associated with your approach, regardless of whether you are analyzing
evidence manually or using machines. Risk is related to uncertainty. Uncertainty
is a potential, unpredictable, and uncontrollable outcome; risk is a consequence of
action taken in spite of uncertainty. The risks we encounter in working with analyt-
ics include risks of project failure and the risk of a failed or poor business decision.
These risks can be immediate or may have long-term consequences. The risks may
fall to the researchers, the business analyst, or decision maker, or they may fall
to the population that is impacted by the research project or the decision maker.
We are just beginning to understand the nature of the risks associated with using
analytical methods, particularly the application of quantitative methods applied to
qualitative sources, and the transformation of qualitative methods from manual to
machine-based. As we gain more experience, it is important to build questions of
risk and uncertainty into our measurement models.

What Level of Investment Are You Willing to Make?

The wisdom of business intelligence methods applies to analytics and particularly
to the new analytical tools. Business intelligence reminds us that there is a trade-
off of time, quality and resources. The new analytical tools require investments of
time and resources, if quality results are to be achieved. One of the reasons business
362 ◾ Analytics and Knowledge Management

analysts and researchers often choose these tools is the belief they will save them
time (e.g., just apply the tools and the solution appears) and require fewer resources
(e.g., the tools will solve all the complex problems). In fact, the tools can require
more computing resources and an extended set of competencies. The tools that have
the greatest value and the most analytical power also come with a high price tag
and maintenance costs. Quality results do not result from lower cost solutions or
lower level resources.

Is This a Project or Enterprise-Level Goal?

We should be clear before we begin a project whether the goal is a one-time effort or
an on-going, enterprise-level operation. In the author’s experience, it is important
to manage expectations. If you never intend to replicate the exercise, if there is no
expectation that the solution has value beyond the immediate application, be clear
about that with your stakeholders. If the solution will be an enterprise application
or one that support continuous operations, it is critical to plan for scale up and
integration. One-time, one-off solutions generally do not integrate well into enter-
prise architecture. They also tend to lack ongoing support from the technology or
design teams. On the research side, it is important for researchers to explain their
methodology and the tools underlying engines and business rules and details for
replicability and reuse.

Understanding Why in Context—Use Case Scenarios

In this section we will walk through the seven use cases and explain how we used
the key questions to guide each of the projects (Tables 12.2 through 12.8). The use
cases are real examples that have been completed over the past 20 years. Each exam-
ple illustrates a different type of goal, a different methodology, different sources of
evidence, and different outcomes. Every use case demonstrates the value of combin-
ing both knowledge management and business analysis strategies. In this section,
we introduce each of the use cases, their goals, the traditional approach to the prob-
lem, measurement strategies, reliability and validity issues, and the risk of errors.
 Analyzing Data and Words—Guiding Principles ◾ 363

Table 12.2 Use Case 1: Causal Relationship between Arab Spring

Revolutions and the Release of Wiki-leaked Cables
Dimension Use Case Context

Primary goals Test the media’s characterization of the relationship

between the content of the Wiki-leaked diplomatic cables
and the motivation for the Arab Spring.

Conceptual Conceptual models of the language of diplomacy and the

models nature of intelligence that is sent from field offices and
diplomatic posts to the Department of State. There are
few theoretical models available given the classification of
the information. The knowledge of subject matter experts
was used to develop models.

Sources of Two primary sources: (1) media reports describing the Arab
evidence uprising and (2) the leaked diplomatic cables. Cables for
two countries were collected—Tunisia and Egypt—for two
years.

Measurement Measurement focused first on the categorization of the

strategy language used in the cables by topics and the
organizations and geographical locations referenced.
Semantic analysis provides the factors for comparison of
diplomatic cables and the media’s characterization of the
causes of the uprisings. Measurement demonstrated that
there was little in little in the cables that was not already
known to citizens of those countries.

Probability of The greatest risk was human perspective and subjective

errors interpretation of the source materials, or the injection of
personal opinion into the analysis. This risk was reduced
by the construction of objective control models of
diplomatic intelligence, and the interpretation of those
models by a neutral machine. Had a control model not
been available, there is a high probability that the full
extent of coverage of intelligence would not have been
exposed.

Level of The initial analysis was research-based and carried little risk
acceptable risk to either the Department of State or the media. The
researcher carried the majority of the risk in the event that
the analysis provided no clear results or produced results
that were not reproducible or which could not be
validated by subject matter experts.

(Continued)
364 ◾ Analytics and Knowledge Management

Table 12.2 (Continued) Use Case 1: Causal Relationship between Arab

Spring Revolutions and the Release of Wiki-leaked Cables
Dimension Use Case Context

Reliability and Because the conceptual model was designed around a full
validity issues view of diplomatic intelligence, it can be reliably applied
to any other country for analysis. Because a conceptual
model was available for control purposes, internal validity
was also strengthened.

Follow-on goals The initial goal was achieved—the research team was able
to demonstrate that a cause-effect relationship of the
leaked diplomatic cables and the uprising did not exist.
Rather, the deep analysis of language and content of the
diplomatic cables suggested that the amount of unique or
novel information in the cables was low when compared
to media reports and social media communication in the
country. This lead to a second research question
comparing the content of diplomatic cables with
in-country media reports.

Table 12.3 Use Case 2: Emotional and Social Tone of the Discourse of
the 2012 U.S. Presidential Campaign
Dimension Use Case Context

Primary goal Characterize the discourse of the 2012 U.S. Presidential

election in terms of its emotional and social nature, to
profile the extensional and intentional nature of the
candidate’s language, and to gauge the density of the
information content.

Existing models Two strong conceptual models existed, including the

Gottschalk-Gleser Model of Emotional Stability and
McLaughlin’s characterization of intentional and
extensional language. The model is available in manual
and automated formats. It gauges levels of emotional
stability, depression, social inclusion, and alienation.
McLaughlin’s model was developed in the 1940s and has
been available only as a manual instrument. In this project
it was transformed into a machine-understandable profile
for semantic analysis. McLaughlin’s model characterized
four indicators of intentional language and four indicators
of extensional language.

(Continued)
 Analyzing Data and Words—Guiding Principles ◾ 365

Table 12.3 (Continued) Use Case 2: Emotional and Social Tone of the
Discourse of the 2012 U.S. Presidential Campaign
Dimension Use Case Context

Sources of Primary sources were speeches and interviews of the

evidence Republican and Democratic candidates for President.

Measurement Semantic analyses produced the information on factors for

strategy analysis. Measurement focused on the variations (e.g.,
central tendencies and dispersion) of factors across
political candidates. The language of the text was first
characterized according to the factors, and variations
across factors were analyzed using quantitative statistics.

Probability of The greatest risk of error would have come from a

errors non-expert characterization of the factors, and in the
subjective interpretation of those factors found in the text
or speech. Risk was reduced by using authoritative
conceptual models. The human subjectivity risk was
reduced through the machine-level interpretation of
those models to the text.

Level of The research goal was best served by a low level of

acceptable risk acceptable risk, given the sensitive nature of the political
process. Risk was managed by identifying and using or
transforming a well-tested and trusted conceptual model.
Both models are accepted by experts in the fields of
psychology, psychiatry, and linguistics.

Reliability and The reliability challenge was in the discovery of sufficiently

validity issues rich texts and speeches from the candidates. The validity
challenge was the possibility that the candidates had not
written the speeches or text, but were delivering thoughts
expressed by others. This was mitigated to a degree by the
inclusion of interview transcripts for each of the
candidates. Another validity challenge that surfaced
through the analysis was the low density of content in
most political speeches and texts.

Follow-on goals The success of the initial semantic analysis of the political
candidates prompted the research team to conduct the
same analysis on media persona from three major
networks. The media texts were contemporaneous to the
candidate’s texts. The follow-on research goal tested the
same factors for media persona and found that in most
cases, the media persona were more extreme than the
political candidates.
366 ◾ Analytics and Knowledge Management

Table 12.4 Use Case 3: Language and Knowledge Structures

of Mathematics
Dimension Use Case Context

Goals Develop insights into how the language used in

mathematical textbooks might contribute to the success or
failure of students learning.

Existing Conceptual models for this project were simple—simple

models linguistic analysis of the use of language in mathematics, and
the characterization of types of knowledge structures (e.g.,
noun phrases) used in each area of mathematics (e.g.,
algebra, geometry, trigonometry, calculus, etc.). This intent
was to develop conceptual models of mathematical language.

Sources of Representative sets of textbooks for each area of

evidence mathematics were selected, deconstructed, OCR’d, and
semantically processed.

Measurement The language of mathematics was characterized through

strategy natural language processing. Grammatical variations were
observed across areas. These variations were offered for
review to subject matter experts.

Probability of The primitive use of natural language processing generated

errors descriptive information about the noun and verb phrases
used in areas of practice. The large volume of text analyzed
for each area was a protection against either type 1 or type II
errors. The analyses generated clear patterns of variations.

Level of As this was exploratory research, we allowed for a high level

acceptable of risk. This research was understood to be the first of
risk several phases. We expected that new hypotheses would
be produced by the findings, and more rigorously tested.

Reliability and Reliability was high for the characterization of language and
validity knowledge structures because each text is grounded in
issues basic principles, assumptions, functions, and symbols. How
language is used to explain concepts or functions may vary,
but the nature and frequency of use of that language is
consistent across texts.

Follow-on A new hypothesis resulted from the exploratory research

goals related to the need to interpret abstract concepts and to
explain mathematical operations in everyday language. We
discovered that while the language of mathematics is intended
to be universal among experts, a layer of interpretation and
translation is required for novices learning the field.
 Analyzing Data and Words—Guiding Principles ◾ 367

Table 12.5 Use Case 4: Knowledge Transfer Practices among Beekeepers

Dimension Use Case Context

Primary goals Understand and characterize knowledge sharing and transfer

among beekeeping communities. Because beekeepers are
one of the most effective and long-lived communities of
practice, we assumed they would be a good source for
study. The intent was to characterize knowledge sharing in
the community (e.g., simple broadcasts, questions and
answers, in-depth discussions, and disagreements).

Existing We began by knowledge modeling of beekeepers

models interactions, and validated the models with expert
members of that community. We used linguistic registers
of actions where they existed.

Sources of Messages from three online beekeeper Listservs were

evidence captured during the most active seasons of the year.
Listserv transcripts were downloaded, linguistically
characterized, and categorized in terms of type of sharing.

Measurement We provided simple measurements of factors that were

strategy semantically modeled to establish a baseline of
understanding.

Probability Because of the primitive linguistic nature of the analysis, the

of errors probability of errors is very low.

Level of This was exploratory research intended to test whether we

acceptable could develop a language-based model of knowledge
risk sharing activities. Risk was borne by the research team.
There was no intervention into a community or its discourse.

Reliability and While the source sample was statistically reliable, the
validity issues conceptual model was applied to only three communities
of beekeepers. The reliability of the research can only be
established for the beekeeping communities. Reliability of
the semantic model of knowledge sharing can be tested in
other communities. Validity of the results is high because
it is grounded on natural language processing and
rule-based linguistics characterizations.

Follow-on The language model that was developed is being developed

goals to other types of communities of practice (e.g., craft
communities). In addition, a new hypothesis has surfaced
for testing—whether the behavior of individuals to share
knowledge and expose errors in an informal setting
changes in a business context.
368 ◾ Analytics and Knowledge Management

Table 12.6 Use Case 5: Precision of Automated Geographical Classification

Dimension Use Case Context

Goals To determine which approach—concept extraction or

categorization—is more effective for determining with a
high level of accuracy what country a text is about.

Existing models Several authoritative models of countries including the Getty

Thesaurus of Geographical Terms, the U.S.G.S.
characterization of geographical entities, and the ISO
country names list. We chose two models—the Getty model
and the ISO country names and synonyms list for testing.
Both models were transformed to machine-based profiles.

Sources of Control set of documents that had been manually indexed

evidence for geographical “aboutness” by human indexers. Same set
of documents were then machine-processed using two
different analytical methods.

Measurement Three way comparison of the accuracy of automated

strategy categorization using the Getty profile, the accuracy of
concept extraction using the ISO model, and with the
manually indexed document as the control point.

Probability of Both type 1 and type 2 errors were noted in the results.
errors However, these errors are not errors of research but errors
of machine-based processes. The fact that the errors were
identifiable and explainable was a significant research
result.

Level of Level of risk depends on the context in which the profile is

acceptable used. If used in the intelligence community, the risk of
risk missing or mischaracterizing a geographical entity can be
significant. If used for general search and discovery, the
risk may take the form of frustration or mistrust—a risk
that may be acceptable.

Reliability and Internal validity of the research was high because of the
validity issues availability and use of the control set of documents.
Reliability may vary, though, depending on the nature of
documents analyzed and of the regional focus of the
documents. The geographical descriptions of some
countries were found to generate higher errors rates than
of other countries.

Follow-on In this case, the follow-on research focused on developing

research deeper understanding of the geographical characteristics
of selected countries.
 Analyzing Data and Words—Guiding Principles ◾ 369

Table 12.7 Use Case 6: Analysis of Physician-Patient Communications

Dimension Use Case Context

Goals Provide a more cost effective and reliable method for

characterizing doctor-patient communication. The
ultimate goal is to provide guidance to doctors on how
to communicate for better health outcomes. Intended
to create a machine-processable tool that can emulate
the human coder at an equally effective or more
effective level in a shorter time.

Existing models Six conceptual models representing the aspects of

doctor-patient communication. All of them require
manual application—human review, interpretation, and
coding. Transformed the factors to machine-based
profiles. Validated the profile with subject matter
experts, remove the subjectivity of application.

Sources of Transcribed text of videotaped conversations between

evidence doctors and patients and doctors and interns.

Measurement Comparison of machine-processable coding to human

strategy coding work. Intent is to continuously improve until we
achieve human level of quality.

Probability of Errors are primarily those of interpretation. Reference

errors manuals describe the nature of the factor in very
general terms. The reduction of errors is dependent
upon our ability to objectively represent the factors and
to represent them in sufficient detail to meet the
expectations or standards of the subject matter experts.

Level of Low level of risk. There is no interference with the actual

acceptable risk medical conversation.

Reliability and Internal validity is an existing problem that we’re trying

validity issues to address through the development of the machine-
based profile. We believe there is inter-coder and
possibly intra-coder variation. Reliability can only be
achieved after the profile has been validated, and tested
on a larger set of interactions.

Follow-on Research to date suggests that there are variations in

research communications among doctors and between doctors
and patients. New factors have surfaced in the focused
analysis of doctor-doctor communication.
370 ◾ Analytics and Knowledge Management

Table 12.8 Use Case 7: Trafficking in Persons Semantic Discovery

Dimension Use Case Context

Goals Identify and model all of the facets of trafficking in

persons, and create a semantic profile that would
enable discovery across the organization, repositories,
locations, and stakeholders.

Existing models Authoritative model exists within the Department of

State but there are variations to be incorporated from
outside the department and by areas of practice.

Sources of Trafficking in persons reports, primary source

evidence materials, formal and informal sources of information,
media reports, law enforcement reports, etc.

Measurement Intent is to improve discovery by creating a single

strategy virtual structure that is easily accessible. Solution must
be comprehensive and inclusive.

Probability of errors Manageable if we work at the facet level—there are

many dimensions of trafficking in persons, each
having its own caveats.

Level of acceptable Low because we are focusing only on access and

risk discovery. If the profile moves beyond this to tracking
communications and discovery actual activities, the
risk would increase significantly.

Reliability and Challenge is to reliability and gaining access to primary

validity issues source materials which are highly classified.

Follow-on research Each facet of the profile can be broken off and used
independently. We would expect further refinement
of pieces of the profile.

How We Use the Tools—Analysis as a Process

After we have defined our business or research goals, levels of investment, measure-
ment strategies, and tolerance for risks and errors, we need to choose an analytical
method. At this stage, it is important that we begin with a good understanding of
analysis. Analysis is a systematic thinking process—a detailed examination of the
elements or structure of something in preparation for discussion, interpretation,
problem solving, or decision making. Historically, there are three basic approaches
to analyzing a situation or thing—quantitative analysis, qualitative analysis, and
a combination of both qualitative–quantitative (e.g., mixed methods). We have
 Analyzing Data and Words—Guiding Principles ◾ 371

a longer tradition of quantitative methods. Qualitative methods have evolved

over the past half-century, and mixed methods are the most recent of the three.
Quantitative methods have the longest history and are grounded in mathematical
and statistical methods. Qualitative methods are more recent and align more with
social sciences and humanities. Mixed methods are the youngest of all. The major-
ity of the analytical tools available today on the commercial market and as open
source tools are statistically or stochastically based. These technologies derive from
and align with quantitative analytical methods. This makes sense because they are
“computer or machine-based” applications.

What Analytical Method Is Best Suited to the Goals?

It is important to understand these analytical methods and how they work.
Applying a good method to an inappropriate or unsuitable source will result in
suboptimal or compromised results. Table 12.9 describes four common analytical
approaches, their functionality and the best-fit sources. The answer to this question
is—it depends on what you’re trying to achieve, what kinds of questions you’re
trying to answer, and what type of source evidence you have to work with.

Quantitative Analysis and Data Analytics

Quantitative analysis is a systematic approach that involves measuring or counting
attributes (i.e. quantities). Quantitative analysis makes use of mathematical mea-
surements and calculations, statistical modeling and research to either describe the
evidence or to infer something from the evidence. Quantitative methods are most
often associated with machine-based strategies because of the complexity of the
calculations and of the volume of data being analyzed. It is simply more efficient
and effective to take a quantitative approach because of the rich foundation of
quantitative methods. In the context of this chapter, quantitative methods trans-
late well to data analytics. Data analytics is supported by a robust methodological

Table 12.9 Domain, Analytical Methods and Source Types

Domain Analytical Method Type of Source

Data analytics Quantitative analysis Quantitative sources

Text analytics Quantitative analysis Qualitative sources

Computational Quantitative and qualitative Qualitative sources

linguistics analyses

Language-based Qualitative analysis Qualitative sources

semantics
372 ◾ Analytics and Knowledge Management

foundation from mathematics and statistics (Han, Pei, & Kamber, 2011; Hand,
Mannila, & Smyth, 2001; Witten, Frank, Hall, & Pai, 2016). There is a robust
market of tools—in the hundreds—to support data analysis, and this market has
been stable for close to two decades. The open source and the commercial markets
are mature so there are options available for organizations of all sizes—from the
simple functions embedded into Excel to the more sophisticated capabilities of
Informatica and SAS. These tools have been used by organizations for decades. The
change in use is in the scale and scope of the evidence they can process and the fact
that they can process live streams of transactional data.
Another type of quantitative analysis that is relevant to this chapter is text ana-
lytics. Text analytics is sometimes also referred to as text mining, text analysis, text
processing, text mining, or text data mining. This type of analysis involves the deri-
vation of information from text—it attempts to infer things from text based on pat-
terns. This approach dates back to the late 1950s and early 1960s and is grounded in
the application of quantitative methods to text (Abbasi & Younis, 2007; Abilhoa &
De Castro, 2014; Arya, Mount, Netanyahu, Silverman, & Wu, 1998; Bejou,
Wrap, & Ingram, 1996; Bengio, 2009; Carletta, 1996; Debortoli, Muller, Junglas, &
vom Brooke, 2016; Deerwester, Dumais, Furnas, Landauer, & Harshman, 1990;
Fernandez et al., 2015; Hofmann, 1999; Lehnert, Soderland, Aronow, Feng, &
Shmueli, 1995; Luhn, 1958a, 1958b; Maarek, Berry, & Kaiser, 1991; Miner, 2012;
Pereira, Tishby, & Lee, 1993; Salton, 1970, 1986; Shmueli-Scheuer, Roitman,
Carmel, Mass, & Konopnicki, 2010; Srinivas & Patnaik, 1994; Teufel & Moens,
2002; Tischer, 2009). Organizations often apply text analytics to a large corpus of
text in an attempt to identify new or interesting patterns or areas of convergence and
similarity. While these methods have continued to evolve over the past 50 years, the
growth has largely been focused on improving the relevance and precision of the
categories or the coverage and recall of targeted entities extracted. In general, these
methods have evolved parallel to, but not with, the evolution of language-based
methods. Text is processed to remove formatting, structure, and language variations
prior to applying quantitative algorithms. Some text analytics tools have natural
language processing (NLP) components that are used to reduce language variations
and surface lemmas and roots. Natural language components are not a function that
is exposed to decision makers or researchers for direct use. The majority of the tools
available on the market today support text analytics solutions.

Qualitative Analysis and Language Based Analytics

Qualitative methods are most frequently associated with manual or human analysis.
There are many types of qualitative research methods and many ways to understand
and apply them. For the purpose of this chapter, we refer to Tesch’s hierarchical
categorization of qualitative research methods because they focus on the elements
that are most closely aligned with knowledge and knowledge management. Tesch
(2013) identified 27 types of qualitative research and organized them into three
 Analyzing Data and Words—Guiding Principles ◾ 373

major substantive questions: (1) What are the characteristics of the language itself?
(2) Can we discover regularities in human experience? (3) Can we comprehend the
meaning of a text or action? Tesch correctly observes that most qualitative analysis
is done with words and focuses on language. Words and language is a very impor-
tant starting point for selecting an analytical method and analytical tools. Typically,
qualitative research tools are designed to facilitate working with information, such as
the conversion or analysis of interviews; the analysis of online surveys; the interpreta-
tion and analysis of focus groups, video recordings, audio records; or the organization
of images. These tools still require human intelligence and effort to support qualitative
analysis. They have little embedded human intelligence. The challenge is that until
recently there were no robust or affordable machine-based tools that automated the
process of analyzing the language of qualitative sources.
One thing we have learned over the years is that a machine can only do high
performance analysis when it has a human-equivalent level of understanding of that
language. Since the 1960s, we’ve devoted considerable research to understanding
how to embed linguistic competence into machine-based applications. Since the
1970s, considerable work has been devoted to building a machine-based under-
standing of language. Computational linguistics is focused on the use of machines
and technologies to support deeper and broader understanding of linguistics and
language. It is a young and interdisciplinary field concerned with the statisti-
cal or rule-based modeling of natural language from a computational perspective
(Anwar, Wang, & Wang, 2006; Berman, 2014; Brandow, Mitze, & Rau, 1995;
Church & Hanks, 1990; Church & Mercer, 1993; Croft, Coupland, Shell, &
Brown, 2013; Eggins, 2004; Hatzivassiloglou & McKeown, 1997; Hearst, 1992;
Hindle & Rooth, 1993; Kaplan & Berman, 2015; Marcus, Marcinkiewicz, &
Santonini, 1993; Mitkov, 2005; Nir & Berman, 2010; Pustejovsky, 1991; Shieber,
Schabes, & Pereira, 1995; Sproat et al., 2001; Van Gijsel, Geeraerts, & Speelman,
2004; Yarowsky, 1995). The theoretical foundations of computational linguistics are
in theoretical linguistics and cognitive science. Applied computational linguistics
focuses on modeling human language use. Since the 1980s, computational linguis-
tics has been a part of the field of artificial intelligence. Its initial development was in
the field of machine translation. The simple automated approach to machine trans-
lation was revisited in favor or more complex development of language and domain-
specific vocabularies, extensive morphological rules engines, and patterns that reflect
the actual use of language in different contexts. Today, computational linguistics
focuses on the modeling of discourse and language patterns. Traditionally, com-
putational linguistics was performed by computer scientists focused on applying
stochastic methods to natural language. In the twenty-first century, though, com-
puter programmers and linguists work together as interdisciplinary teams. Linguists
provide critical knowledge of language structure, meaning, and use that guide the
use of analytical tools. Because these tools have an academic or research focus, com-
putational tools are often open source or laboratory-based. By and large, they are not
available on the commercial market.
374 ◾ Analytics and Knowledge Management

Natural language programming is a critical component of today’s qualita-

tive analysis methods and tools. It focuses on the interactions between human
language and machines. It is an interdisciplinary field that lives primarily in
computer science but includes major contributions from linguistics, artificial
intelligence, and cognitive science (Ballard & Biermann, 1979; Biermann, 1981;
Biermann, Ballard, & Sigmon, 1983; Chowdhury, 2003; Dahl & Saint-Dizier,
1985; Dijkstra, 1979; Halpern, 1996; Heidom, 1971; Jurafsky, 1996; Liddy, 2001;
Loper & Bird, 2002; Manning & Schutze, 1999; Marcu, 1997; Mihalcea, Liu, &
Lieberman, 2006; Spyns, 1996; Tomita, 2013; Weizenbaum, 1986). Over the
past 50 years, computer programmers, linguists, and information scientists have
worked together to create these core components. We have constructed machine-
based components to support all of the levels of language. NLP applications must
be developed for each language, and for geographical variations of languages. The
most important element of an NLP application for analysts and knowledge scien-
tists is the part of speech (POS) tagger. In order for a POS tagger to work, we need
two things: an extensive characterization of the target language and a framework
or set of tags. The tag set should include all possible combinations of category val-
ues for a given language. A tag set is generally represented by a string of letters or
digits. As with computational linguistics tools, natural language programming and
processing tools are generally used in academia by researchers. NLP is a key tool
used by computational linguists, and computational linguists have been critical to
the development of these methods.
Language-based semantic methods and tools is the youngest of all and per-
haps the least understood of all the terms. Language-based semantic methods
are a combination of all of the above concepts and capabilities. Language and
knowledge-based applications typically leverage these components but in a designer
or configurable way. These applications are like toolboxes—they contain core lan-
guage and knowledge modeling tools and allow users to build solutions to suit
different business or research problems. Among the tools in a semantic analysis
toolbox we would find NLP and natural language programming, POS taggers,
concept and entity extraction tools, grammar and rule-based categorization con-
struction tools, indexing tools, abstracting rule builders, and clustering engines.
The capabilities are often beyond the current competencies of business, so they do
not receive a significant treatment in the popular media or even in the scholarly
press (Edmundson, 1969; Goddard, 2011). These methods are rarely all found in
commercial tools, though there are a few examples.

Mixed Methods Analysis and Variant Sources

Mixed methods analysis involves collecting, analyzing, and integrating quan-
titative and qualitative methods and sources. In essence, this type of analy-
sis takes advantage of using multiple ways to explore a research problem. We
 Analyzing Data and Words—Guiding Principles ◾ 375

may begin by using linguistic registers and knowledge modeling to understand

the content of text corpora, then switch to quantitative classification models to
model what we have learned, and switch back to semantic methods to classify
the full corpus. The mixed methods approach is a fairly young discipline—
such methods have only been accepted by the research community for the last
decade. When using a mixed methods approach, the rigor of the design is criti-
cal. Mixed methods are not the simple application of quantitative methods to
qualitative sources.

When Is a Quantitative Analysis Approach Warranted?

Quantitative analysis is appropriate when the source evidence is quantitative or
structured in nature, where the volume of structured data is significant, or when
the quantitative variables are complex, and when the intent is to generalize the
results to a larger base or population. We generally know something about the data
under analysis when using quantitative methods. We would not apply a quantita-
tive tool to “messy” or unmanaged data. Using standard methods means that the
research can be replicated, analyzed, and compared with similar studies (Kruger,
2003). One of the attractions of quantitative methods is that they minimize the
personal bias that is inherent to qualitative methods—the machine is doing the
analysis following predictable rules.

When Is a Qualitative Analysis Approach Warranted?

Qualitative analysis is appropriate when we are working with language or text,
when we are interested in learning the characteristics of the language, when we are
interested in discovering regularities in human experience, and when we are trying
to discern meaning of text, language, or action. Qualitative methods are closely
aligned with the focus and interests of knowledge sciences—sense-making, under-
standing, deriving meaning. They are solidly anchored in the use of language and
the analysis of text and speech.

When Should We Choose a Mixed Methods Approach?

A mixed methods analysis is particularly well-suited for contexts where we are try-
ing to develop a better understanding of a research or a business problem, where
interpretation of the problem is continuously evolving, where sample sizes may vary
depending on what we are trying to learn, and where we may need to leverage more
than one technique to collect source evidence. We used mixed methods to explain
and interpret, to explore, to strengthen a single design strategy, to overcome the
weakness of a particular design, and to generally look at a research problem from
multiple levels.
376 ◾ Analytics and Knowledge Management

Which of These Analytical Methods Are Supported by Tools?

The challenge with the market today is that there is no clear description of the func-
tions that the tools support—the types of analyses that we can perform. Tools are
described with words like “analytics,” “semantics,” “data to text,” “discovery,” and
“language tools” without any real explanations of what they are and what they do.
The only way to learn what you’re getting is to have a good understanding of what
you need, and a set of sample test cases to use to determine what any given tool does.
Ideally, we should be able to identify tools that will align well with quantitative analy-
ses and quantitative sources, tools that align well with qualitative analyses and quali-
tative sources, and a variety of tools that support different stages of a more complex
mixed methods analysis. We need to understand how the tool treats words (e.g., are
they reduced to data points or are the words treated with language tools?), how the
tool prepares data for analysis (e.g., is there a set of routines that removes formatting,
stop words, etc.?), what the analytical engine does (e.g., does it look for patterns, deter-
mine frequencies and associations, or are there multiple tools that can be “designed
into a solution?”), is there an underlying dictionary and how is it used (e.g., what is the
source—Wall Street Journal vocabulary, Oxford English Dictionary, etc.?).
In addition to the basic functionality of the tool, we need to understand whether
the tools bundle products (e.g., they generate a search index but do not expose the
results to developers, they create and assign categories but do not enable direct
definition of categories, etc.), whether the tool is designed to work with particular
content (e.g., social media, text messages, words in spreadsheets or databases, or
large text documents), and products that intentionally blend the analysis of data
and text without treating either according to its basic characteristics or behaviors.
It is critical to know which analytical methods are supported and how they align
with your business needs.

Which of These Analytical Methods

Are Not Supported by Tools?
It is not surprising that the majority of the analytical tools today are quantitative
because it has taken many more years to develop the machine readable and under-
standable linguistic foundations that are needed to support qualitative source anal-
yses. And, this foundation is more complex than simple statistical or mathematical
methods. In fact, analyzing language will generally require the use of multiple
qualitative tools—POS taggers, linguistic register generators, knowledge model-
ing tools, entity extraction tools, categorization scheme tool builders, automated
indexing tools, and tools that allow us to specify abstracting and indexing rules.
There is no one tool that supports all of the steps in a qualitative analysis simply
because each of these qualitative analyses methods works from a different set of
processes and rules. Consider how a human approaches and thinks about each of
these tasks, and what we need to teach a machine before it can perform the same
 Analyzing Data and Words—Guiding Principles ◾ 377

functions. Other key tasks involved in qualitative analysis such as knowledge elici-
tation, knowledge modeling, and knowledge representation are inherently human.

What Opportunities Are There for Mixing Analytical Methods?

The opportunities for supporting mixed analytical methods are significant but
they are dependent upon good design. Understanding where and when we can
use quantitative and qualitative methods in our design is the key question. In fact,
every qualitative analysis will be a mixed methods solution. We may begin with an
analysis and characterization of language, extract knowledge structures, and then
use quantitative methods to characterize the use of those structures in a corpus.

How in Context—Use Cases

In this section, we consider these questions as they apply to the seven use cases
(Tables 12.10 through 12.16). Choices made at this point can mean the difference
between success and failure.

Table 12.10 Use Case 1: Exploring the Causal Relationship between Arab
Spring Revolutions and Wiki-leaked Cables
Dimension Use Case Context

Design choices Multistep design process that focused on the topical

categorization of the content, references to
geographical and organizational entities, and
concept extraction of media texts.

Quantitative methods Quantitative methods could be applied to the entities

and categories generated through semantic analysis.
Quantitative data analytics, though, were not
appropriate for characterizing the language content.
Research studies which took this approach reported
achieving no significant results.

Qualitative methods Majority of the initial preparation work was

qualitative in nature—focusing on the topical
description of the corpus, and the identification of
targeted entities. The qualitative analysis was more
rigorous through the intentional and objective
development of semantic profiles based on the
knowledge models of experts.

Mixed methods Mixed methods were the preferred approach. Neither

qualitative nor quantitative methods would have
sufficed.
378 ◾ Analytics and Knowledge Management

Table 12.11 Use Case 2: Emotional and Social Tone of the Discourse of the
2012 U.S. Presidential Campaign
Dimension Use Case Context

Design choices Because this project had so many different

dimensions and two distinct populations—media
and political candidates—a multi-step design
process was required. This approach worked well
because it allowed us to pause and check the
results of each step.

Quantitative methods Quantitative methods were used for the final

analysis—to formulate scales of impact and to
determine the significance of variations.

Qualitative methods Knowledge models of factors were critical to this

project. Expert models of psychological stability,
reference models of language use, and concept
identification and modeling were all critical to the
analysis.

Mixed methods The entire project should be seen as a mixed

methods analysis.

Table 12.12 Use Case 3: Language and Knowledge Structures of

Mathematics
Dimension Use Case Context

Design choices Primary focus was on the nature of language

(linguistic register) and the knowledge structures.
It was clear that there was variation in both of
these by area of mathematics—it is not one
universal language. Stepwise design was also
critical here.

Quantitative methods Quantitative methods were only used to describe

the results and to analyze the variations across
areas of practice.

Qualitative methods Primary focus was on profiling language and

extracting knowledge structures.

Mixed methods Primarily a qualitative research analysis, but with

some use of quantitative methods in the final
stage.
 Analyzing Data and Words—Guiding Principles ◾ 379

Table 12.13 Use Case 4: Knowledge Transfer Practices among Beekeepers

Dimension Use Case Context

Design choices Primary focus was on the use of language in

beekeeper discourse, so we began by building a
linguistic register. Factors pertaining to knowledge
transfer were identified, modeled and tested.

Quantitative methods Quantitative methods were only used to describe the

basic population and the nature of the interactions
(i.e., communication networks), and to analyze the
significance of the factors surfaced through
semantic analysis.

Qualitative methods Knowledge structures were identified from

authoritative sources; models were constructed and
built into profiles. Profiles were applied to each
thread and each list entry.

Mixed methods Quantitative methods were used to characterize and

validate the sample population and to perform the
final quantitative analysis on the results of the
semantic analysis. This was a mixed methods
analysis.

Table 12.14 Use Case 5: Precision of Automated Geographical Classification

Dimension Use Case Context

Design choices Two strategies were built into the design, one
focusing on categorization profiles and the other
focusing on text analytics. Two parallel designs
were run simultaneously and then quantitatively
tested in the final stage.

Quantitative methods Quantitative methods were used to support the

text analytics strategy, and they were used in the
final stage to compare the results of both design
strategies..

Qualitative methods Qualitative methods were used to model and

construct the country based categorization
profile. Qualitative methods were also used to
align the model with the expert profile developed
by the Getty Institute.

(Continued)
380 ◾ Analytics and Knowledge Management

Table 12.14 (Continued) Use Case 5: Precision of Automated Geographical

Classification
Dimension Use Case Context

Mixed methods The categorization profile leveraged mixed

methods, though the text analysis strategy
leveraged only quantitative methods. One strategy
was not an example of mixed methods, but the
entire project as a whole was.

Table 12.15 Use Case 6: Analysis of Physician-Patient Communications

Dimension Use Case Context

Design choices The design is a machine-transformation of existing

manually applied profiles. Six profiles were found
in the literature, and each was interpreted and
transformed for machine processing. This
required first generating knowledge models of
each factor from each profile, then developing the
rules to represent that factor.

Quantitative methods The only quantitative methods are those that

measure the thresholds established for
categorization. We have profiles to work with but
we do not have access to highly confidential
manually coded conversations from other
researchers. There is not a control group for us to
work with.

Qualitative methods The majority of the analysis is grounded on

qualitative methods used to create profiles of
medical and health language.

Mixed methods The majority of the analysis here was qualitative

rather than quantitative. Quantitative methods
would have introduced additional risk, and risk
would not have been acceptable for tagging
future conversations with a machine-based
profile.
 Analyzing Data and Words—Guiding Principles ◾ 381

Table 12.16 Use Case 7: Trafficking in Persons Semantic Discovery

Dimension Use Case Context

Design choices Two methods were used—text analytics and

semantic categorization. The intent was to
compare and contrast the results, and to identify
the best fit role for each in the organization’s
information management and discovery strategy.
Two parallel analyses were implemented and
compared in the final step.

Quantitative methods Quantitative methods in the form of stochastic

clustering were used for one track. The results
were not well suited to integration and discovery
in a dynamic environment and across different
types of language and domains.

Qualitative methods The semantic profile was constructed using

concept extraction and knowledge modeling
methods. The profile was labor and consultation
intense to construct but it is sustainable over
time.

Mixed methods Semantic categorization profile strategy was

mixed methods. Text analytics strategy was
largely quantitative. In total, the project was
mixed methods.

What Sources Are We Analyzing?

It is important to distinguish between quantitative data or evidence (repre-
sented as numerical or structured and interpreted) and qualitative data or evi-
dence (expressed through human language). Quantitative sources and qualitative
sources are fundamentally different in their structures, interpretation, meaning,
and behaviors. Reducing language to data or digits removes context and meaning.
Adding meaning back into the results after the data processing and cleansing step
of text analytics introduces additional subjectivity that can either compromise the
interpretation or render the results invalid. Human knowledge and understand-
ing is still the best source of evidence we have to support analysis. This is the
area of greatest potential and opportunity for knowledge organizations. It is also
the dimension with the greatest threat of failure. How will the machine treat or
process the language? It is not sufficient to build common language dictionaries
and stemming algorithms into a product and assume that this means the machine
will understand the language. We need to stop and ask what components need
382 ◾ Analytics and Knowledge Management

to be designed into the application for a machine to process language as well as a

human. These components are available today but they are not always present in a
“text analytics” tool.

What Is the Nature of Language, Information,

and Knowledge We’re Using as Source Evidence?
Text, speech, and knowledge are all qualitative sources of evidence. Qualitative
sources of evidence are expressed in words. Words have meaning and are embedded
into linguistic structures that support the language of expression. Words cannot
and should not be treated as “data points” without a full understanding of the
language. Reducing language to make it palatable for quantitative analysis does
not surface or expose meaning—it masks and buries it. Language is essential to
understanding meaning, to knowledge transfer, to teaching machines how to com-
municate, understand, and manipulate knowledge. Any analysis of text, speech or
expressed thoughts must begin with an understanding of the underlying language
patterns and variations.
Not all languages have the same linguistic structures. Not all uses of lan-
guage are comparable even within the same language. The context in which
language is used varies for a single individual, across communities, across disci-
plines, and across social classes. Humans intuitively understand these variations
and bring that understanding to any manually performed qualitative analysis.
When we ask a machine to perform that same type of analysis, the machine must
be grounded on a foundation of language understanding. No computer-based
analysis of text, speech, or expressed thoughts can be valid unless it takes into
consideration the basic elements of language. To humans, language looks simple
because after we have learned it we no longer think about its construction and
use. We just use it. But language is complex. Humans doing qualitative analysis
understand language—machines doing qualitative analysis must be able to attain
a similar level of understanding if we are to trust their results. There are five levels
(Figure 12.1) that explicitly and formally describe a language. Machines must be
able to understand and process all five levels. The five levels are phonetics, phonol-
ogy, morphology, syntax, and semantics.

Level 1: Phonetics represents the most primitive signal level—whether

expressed as text or acoustic signals. Phonetics is generally represented in a
language’s phonetic alphabet. This is the level at which the machine begins
to recognize language—it understands symbols. Without this level of rec-
ognition there is nothing on which to build. An important system level
issue here is representation of symbols in Unicode. If the tool you are using
does not require text in Unicode, you are not working with a language-
based tool.
 Analyzing Data and Words—Guiding Principles ◾ 383

Level 1: Phonetics

Level 2: Phonology

Level 3: Morphology

Level 4: Syntax

Level 5: Semantics

Figure 12.1 Five levels of language.

Level 2: Phonology is a pattern or sequence of phones (or sounds) or text patterns

sequence of letters. This is the first level at which we have formal patterns of
sounds and symbol patterns. This level of understanding is also essential for
audio and speech recognition, for translating sound into text patterns, and
for smart stemming.
Level 3: Morphology deals with the building of word forms from phonemes and
their lemmas. Morphology helps us with dictionary lookup by identifying
dictionary headwords, and with and understanding its grammars. Every lan-
guage has morphological rules and behaviors—such as inflection, derivation,
compounding, and so forth. This level of language understanding helps the
computer to understand some of the rules for building and also deconstruct-
ing words in a particular language. It is not simply a matter of the machine
identifying nouns and verbs, but determining the tense of the verb, whether
the noun is singular or plural, and the nature of the adjective. All of this
knowledge—which humans bring to a manual qualitative analysis task—
must be engineered into computer-level understanding.
Level 4: Syntax is the acceptable structure of a sentence in a language. Syntax
is important because it is the first level at which the machine begins to
understand the context of the language we wish to analyze. Syntax also is
the level at which the complexity of language becomes evident. We think
of syntax in terms of open and closed POS. Closed categories typically do
not change—they do not decline. Closed categories include preposition,
conjunction, article, interjection, and particles. This is a limited (finite) set
of words. It is easier to build these into a language base. Prepositions (of,
without, by, to), coordinating conjunctions (and, but, or, however), subor-
dinating conjunctions (that, if, because, before, after, although, as), articles
or determiners (a, an, the), interjection (wow, eh, hello), and particles (yes,
no, not). Open POS typically can change—through inflection, declension,
384 ◾ Analytics and Knowledge Management

newly created words, transforming words. Open categories include verbs,

nouns, pronouns, adjectives, and adverbs—each important to analyzing
language. For example, the investigation of conditional statements will rely
heavily on adverbs. The exploration of hyperbolic speech or superlatives
will focus on adjectives. Verbs form the key representation of actions or
operations—for example, we would focus on verbs to study medical advice
or a team’s activities. If we cannot isolate and explore verbs, we cannot effec-
tively portray how a call center is advising callers. It is not sufficient to work
simply with the roots or lemmas of these words—in some cases the tense or
the declension is a critical aspect of the analysis. Working with a tool that
simply stems, matches letters, and looks for patterns among words does
not support this level of exploration. Reducing language to “data points”
may simplify language for processing, but in so doing we lose the meaning
and context—understanding.

What Meaning or Understanding Are We Deriving

from the Language?
Semantics is level 5 in the characterization of language. This is the highest level of
language understanding and representation. Computer applications that purport
to understand and interpret “text” or “speech” must have the building blocks of
levels 1 through 4. There can be no derivation of meaning from text, speech, or
any other expression of language without morphological (level 3) and syntactical
(level 4) analyses. It is at level 5 that humans begin to teach the machine what we
understand and what we know. We do this by modeling knowledge and translating
what we know intuitively into linguistic registers and knowledge structures, and by
constructing knowledge-based processes.

What Linguistic Registers Are Represented

in the Source Evidence?
Language varies by social context, by economic class, and by subject discipline.
Most of the work on linguistic registers is systematic qualitative research by lin-
guists. A linguistic register is a set of specialized vocabulary and preferred (or
unpreferred) syntactic and rhetorical structures used by specific socioprofes-
sional groups for special purposes. Bar-Hillel (1966), Barton and Hamilton
(2012), Berman (2014), Berman and Nir (2009), Berman and Verhoeven (2002),
Childton (2004), Cope and Kalantzis (2000), Eggins (2004), Eggins and Slade
(2005), Fairclough (2003, 2013), Ferguson (1983), Finegan and Biber (1994),
Gee (2015), Halliday, McIntosh, and Strevens (1964), Henzl (1973), Kaplan
and Berman (2015), Kress (2009), Neumann (2014), Ravid and Berrman (2009,
2010), Schiffman (1997), Spigelman (1999), and Ure (1982) were among the first
 Analyzing Data and Words—Guiding Principles ◾ 385

to address the idea of registers. Linguistic registers represent a person’s ability to

adapt their use of language to conform to standards or traditions in a given profes-
sional or social situation. The human-equivalent level of understanding of a con-
text or domain looks exactly like a fully-elaborated linguistic register. Over their
lifetime, humans develop and use a repertoire of linguistic registers to use in social
contexts, in work environments, when talking with others in a subject domain,
and when expressing emotions or spiritual thoughts. Before we can undertake an
analysis of text or a corpus of text, we must understand the embedded linguistic
registers.
Machines will not be able to distinguish these variations with their limited
understanding of language unless we can teach them how to distinguish. To build
out linguistic registers for major domains may take us as long as it did to build some
of the semantic foundations of our current applications and technologies. It is well
worth the effort, though, because this is the key to really building more intelligent
semantic applications. The register of law is different from the register of medi-
cine, which is different from the language of engineering and from the language of
mathematics.

What Knowledge Structures Are Represented

in the Source Evidence?
Critical components of any linguistic register are the structures used to represent
knowledge. Each person who uses language to communicate, to share ideas or
to record information is drawing upon a rich repertoire of knowledge structures.
Analyzing text without understanding the linguistic registers or knowledge struc-
tures is comparable to taking uncontrolled sources of data, throwing them all
into a repository without any data management effort, and expecting statistical
applications to sort out all of the data variations and discrepancies for us. At
the highest level, we develop models that represent knowledge structures, includ-
ing declarative knowledge, procedural knowledge, and structural knowledge.
Declarative knowledge tells us how and why things work the way they do, names,
locations, concepts, elements, and relationships among concepts. Procedural
knowledge describes the steps or activities required to perform a task, processes,
and actions. Structural knowledge is used for plans, strategies, problem solving,
critical thinking, and so on. At the lowest or most primitive level, though, we
are talking about POS. All of these are important to creating a linguistic register
for a context or domain. We begin to identify knowledge structures, though,
by looking for tendencies and patters at a very primitive level; by identifying all
of the noun and verb phrases that are commonly used. Elements of knowledge
structures include concept names and variations on those names, definitions and
concepts, related or associated concepts, and elaborated descriptions of the rela-
tionships among concepts.
386 ◾ Analytics and Knowledge Management

What Types of Semantic Methods Are Available

for Us to Work With?
Almost every analytical tool on the market today is described as “semantic.” In
fact, these products may not be semantic as we’ve defined it but data analytics, text
analytics, or simple NLP applications. The important issue here is to know what
you need and what the tool you’ve chosen is capable of doing.

What Does the Analytical Workflow Look

Like for Each of These Methods?
Let’s compare and contrast the workflow models for text analytics and for
language-based applications. Text analytics generally follows a three step pro-
cess: (1) text acquisition and preparation—identifying the source evidence and
performing preliminary formatting of the text; (2) making the text readable
at the computer level by producing optical character recognition (OCR)-ready
text, removing stop words, parsing and chunking text into meaningful “data
points”; and (3) applying statistical methods such as Bayesian statistics or
k-nearest neighbor to identify patterns within the text. In contrast, linguistic or
language-focused analysis begins with a statement of goals and requirements,
considers the type of solution needed, proceeds to testing and ultimately to
deployment. This strategy focuses on understanding the source evidence and
aligning the design with the goals. Knowledge modeling and representation,
as well as consultation with subject matter experts, and existing theoretical
models are important elements of this analytical approach. The goal is to teach
the computer what we are looking for and how a human would think about it.
This is in contrast to focusing on the text, preparing the text for data analysis,
and then applying the analytical methods.

Do These Methods Offer the Possibility of Reducing

or Eliminating the Subjective Interpretation Element?
Reducing subjectivity and increasing objectivity should be a primary goal whenever
we are working with qualitative sources of evidence. Our ability to achieve this
goal, though, is dependent upon our ability to objectively model the problem we
are exploring and the knowledge we are using as source evidence. Any new refer-
ence or conceptual models we construct of factors should be derived from the work
of experts in those fields—wherever it exists. In addition, any models we construct
should be validated with subject matter experts before they are applied to text. We
reduce subjectivity by exposing our assumptions and constructions at each step of
the analytical process.
 Analyzing Data and Words—Guiding Principles ◾ 387

What Is the Return on Investment for the Analysis?

The return should be greater than the investment. This may sound simple and logi-
cal, but it is easy to become enamored with the tools and to overinvest in an analy-
sis. While the marketing and promotional materials lead us to believe that every
application will be cost effective, these materials do not take into consideration the
setup, configuration, modeling, training, and operational support the tools require.
Where there is a good understanding of the level of investment required and where
the return is greater than the investment, we have a good opportunity.

What Level of Effort Is Involved in Doing

a Rigorous Analysis?
The answer to this question is—it depends. Some applications can take up to six
months and require one or more fulltime employees. The scale of the effort depends
on the complexity of the design model, the complexity of the language, and the
scale of the application. It is important to understand the organization’s tolerance
for or expectation of resources required. Many organizations may expect little or no
investment based on the promotional materials. A general rule of thumb is that if
effort and thought are invested in the development of the model and methods, the
level of effort to apply will be reduced.

What Types of Competencies Are Needed

to Support the Analysis?
The competencies required to work with qualitative sources, to model and charac-
terize language, and to develop machine-based solutions include linguistic meth-
ods, linguistic registers, knowledge modeling and structure methods, knowledge
engineering, knowledge elicitation, natural language programming, statistics,
general research methods, subject matter expertise, and knowledge and informa-
tion organization methods. Any successful application clearly requires a team
effort.

What Is the Feasibility of the Analysis

Without Semantic Methods?
None of the seven use cases we have presented would have been feasible or economi-
cally practical without semantic methods. While semantic methods may not have
been the only method we used, it played a critical role in each use case. It will be
the case that any analysis that leverages language-based source evidence requires
the use of semantic methods at some point in the design.
388 ◾ Analytics and Knowledge Management

Understanding Sources in Context—Use Case Examples

As the use case explanations suggest, the most important thinking point is the
nature of the source evidence and our characterization and representation of lan-
guage. The key questions pertaining to sources are considered in the context of the
seven use cases (Tables 12.17 through 12.24).

Table 12.17 Use Case 1: Causal Relationship between Arab Spring

Revolutions and the Release of Wiki-leaked Cables
Dimension Use Case Context

Nature of information Diplomatic cables from Tunisia and Egypt, media

and knowledge reports of the Arab spring, other treatments and
discussions of the nature of diplomatic language.

Linguistic register(s) Yes, linguistic registers were critical to

understanding and validating the nature of
diplomatic language, as well as the language of
media reports.

Knowledge structures It was important to expose and describe the

types of knowledge structures included in
diplomatic cables. We found people,
organizations, locations, events, and topics.

Semantic methods used Concept extraction, entity extraction,

categorization

Design workflow Language-focused workflow

Competencies needed Knowledge of diplomacy, knowledge of the

foreign policy interests in the countries,
linguistic competencies, understanding of
media reporting and language.

Return on investment Very high in the longer term if the profile can be
used to model and assess the intelligence
information that is being sent back to
headquarters. If the incoming intelligence is
equivalent to media reports from the country,
there is a need to improve foreign intelligence
gathering work.

Qualitative risk reduction Because of the nature of diplomatic language,

there is little emotion or personal perspective
expressed. This language carries a lower risk for
analysis.

(Continued)
 Analyzing Data and Words—Guiding Principles ◾ 389

Table 12.17 (Continued) Use Case 1: Causal Relationship between Arab

Spring Revolutions and the Release of Wiki-leaked Cables
Dimension Use Case Context

Feasibility without The multifaceted nature of the semantic profile

semantic tools suggests it would take a very long time to
complete the analysis. Also suggests that
manually operationalizing the analysis in the
future is not practical.

Table 12.18 Use Case 2: Emotional and Social Tone of the Discourse of the
2012 U.S. Presidential Campaign
Dimension Use Case Context

Nature of information Speech transcripts from candidates, media

and knowledge transcripts of television programs.

Linguistic register(s) Linguistic registers were generated for individual

test populations, and for the full set.

Knowledge structures Knowledge structures were developed around

the expertise of psychologists and psychiatrists.
Entity based knowledge structures were used to
examine the density of information in the
language of both groups.

Semantic methods used Concept extraction, categorization

Design workflow Language based workflow was used.

Competencies needed Subject matter expertise in area of emotional and

social stability defined by psychiatrists and
psychologists; linguistic competencies,
knowledge modeling, categorization scheme
construction, entity extraction.

Return on investment The return on investment was significant because

it clearly explained and supported the
depressive nature of media and political
discourse. It also provides a profile for analyzing
speech as it is delivered.

Qualitative risk reduction Because the project leveraged semantic profiles

developed by psychologists and psychiatrists the
qualitative control points are strong.

(Continued)
390 ◾ Analytics and Knowledge Management

Table 12.18 (Continued) Use Case 2: Emotional and Social Tone of the
Discourse of the 2012 U.S. Presidential Campaign
Dimension Use Case Context

Feasibility without Without the machine-based tools this project

semantic tools would not have been possible or feasible. The
complexity and labor intensive nature of the
analysis would have been cost prohibitive. In
addition, without the expert profiles, the degree
of subjective interpretation would have been
significant.

Table 12.19 Use Case 3: Language and Knowledge Structures

of Mathematics
Dimension Use Case Context

Nature of information Mathematical textbooks, mathematical

and knowledge instruction materials, mathematical research
papers in algebra, geometry, trigonometry,
advanced algebra, calculus, discrete
mathematics, industrial mathematics, and
general advanced research topics.

Linguistic register(s) Linguistic registers were generated to understand

the use of verbs in particular as indications of
how operations and functions are explained.

Knowledge structures Knowledge structures were built around noun

phrases to represent the basic concepts and
objects addressed in each area of mathematics.

Semantic methods used Simple part of speech tagging, simple concept

extraction.

Design workflow Language focused workflow was followed.

Competencies needed Basic competencies such as language modeling,

and interpretation of verb phrases. Knowledge
of mathematics was also important.

Return on investment In the longer term, the return can be significant if

the conclusions and findings are adopted by
teachers of mathematics and writers of
mathematical textbooks.

(Continued)
 Analyzing Data and Words—Guiding Principles ◾ 391

Table 12.19 (Continued) Use Case 3: Language and Knowledge Structures

of Mathematics
Dimension Use Case Context

Qualitative risk reduction Because there is a high level of fear of

mathematics among the general publication,
managing the risk of subjective interpretation
was important. The fundamental nature of the
analysis kept this risk to a minimum.

Feasibility without The intensity and the simple density of the

semantic tools language of mathematical textbooks make this
project cost prohibitive and labor intensive
without machine-based methods.

Table 12.20 Use Case 4: Knowledge Transfer Practices among Beekeepers

Dimension Use Case Context

Nature of information Listserv conversations captured as threads over a

and knowledge 3-month period. Three different Listservs were
monitored during the most active beekeeping
season of the year.

Linguistic register(s) Linguistic registers were created to understand

the language of beekeeping. The register was
compared to the language used in beekeeping
training materials, beekeeping textbooks, and
beekeeping course materials. Also analyzed was
the language of knowledge transfer, sharing, and
failure exchange.

Knowledge structures Knowledge structures were extracted and

compared to those used in primary and
secondary source materials.

Semantic methods used Language modeling, part of speech tagging,

semantic categorization of knowledge transfer,
and failure exchange.

Design workflow Language modeling workflow was followed.

Competencies needed Practical knowledge of beekeeping, linguistic

knowledge, semantic analysis strategies.

(Continued)
392 ◾ Analytics and Knowledge Management

Table 12.20 (Continued) Use Case 4: Knowledge Transfer Practices among

Beekeepers
Dimension Use Case Context

Return on investment The return will be realized by organizations as the

research moves forward. The study explores the
knowledge sharing behavior outside of formal
organizations. The next step in this research is to
determine how those behaviors change when
the context changes and individuals no longer
have a direct stake in success.

Qualitative risk reduction There is a risk of subjective interpretation

inherent to all knowledge sharing studies. By
focusing on language and neutralizing the
subjective elements through profile
development, there is an opportunity to reduce
risks.

Feasibility without The volume of data could have been managed

semantic tools manually. However, the subjective interpretation
of discussion of errors and knowledge sharing
exchanges would have presented challenges for
the validity of the research.

Table 12.21 Use Case 5: Precision of Automated Geographical Classification

Dimension Use Case Context

Nature of information Internal documents of various types from an

and knowledge international development organization.
Documents had been country classified
manually to provide a robust control group.
Documents and their metadata were used in the
research analysis.

Linguistic register(s) Registers were not created because we focused

on a single entity—country.

Knowledge structures Two knowledge structures were developed—one

a simple representation of country names and
variations of those names, and the other a fully
elaborated country classification that included
all geographical features, the nationalities and
ethnic groups in the country, names of all cities,
and administrative jurisdictions.

(Continued)
 Analyzing Data and Words—Guiding Principles ◾ 393

Table 12.21 (Continued) Use Case 5: Precision of Automated Geographical

Classification
Dimension Use Case Context

Semantic methods used Simple entity extraction, rule-based categorization.

Design workflow Language-focused workflow was followed for the

second strategy, and simple text analytics
workflow was followed for the first strategy.

Competencies needed Knowledge of geographical descriptions,

categorization, knowledge of authoritative sources.

Return on investment Potential return is significant because this is

currently a process which is prone to high error
rates and to high costs from manual work.

Qualitative risk reduction The risk is of type 1 and type 2 errors, but not
associated with subjective human interpretation
of source materials.

Feasibility without Without machine-based methods, it would have

semantic tools been cost prohibitive to test the categorization
profile. The complexity and number of match
points in the profile is too large for humans to
compare against individual documents.

Table 12.22 Use Case 6: Analysis of Physician-Patient Communications

Dimension Use Case Context

Nature of information Videotaped conversations of doctors, interns,

and knowledge and patients which were manually transcribed
by graduate students.

Linguistic register(s) Linguistic registers were created to understand

the language of communication between these
different stakeholders. Part of speech tagging
was critical to this project.

Knowledge structures Knowledge structures were defined through

reference to the six expert profiles found in the
literature. Concept extraction was used to
identify factors highlighted in the profiles.

Semantic methods used Part of speech tagging, concept extraction,

categorization.

(Continued)
394 ◾ Analytics and Knowledge Management

Table 12.22 (Continued) Use Case 6: Analysis of Physician-Patient

Communications
Dimension Use Case Context

Design workflow Language-focused workflow.

Competencies needed Communications expertise, medical and health

language expertise, semantic analysis methods.

Return on investment Return could be significant in terms of scaling up

the profiling of existing communications, and
transforming what we have learned to teaching
methods for medical students and doctors.

Qualitative risk reduction Subject risk for this area is very high depending
on one’s role and perspective. An objective
analysis of the discourse can identify ways that
communication can be improved. There is a
potential to impact health outcomes by
improving communication and trust between
doctor and patient.

Feasibility without The risk of subjective interpretation is high, and

semantic tools this risk can impact the trust and validity of any
research based on manual coding.

Table 12.23 Use Case 7: Trafficking in Persons Semantic Profile

Dimension Use Case Context

Nature of information Department of State trafficking in persons annual

and knowledge reports, discussions of facets of trafficking in
research papers, in law enforcement, in health
services, and in economic contexts. Leveraged
reports and language from other organizations
also working in this area such as the UN, ILO,
WHO, USAID, and so on.

Linguistic register(s) Linguistics registers were generated to better

understand the variations in language used
across stakeholder groups.

Knowledge structures Knowledge structures are primarily focused on

noun phrases; however, the profile also exposes
attitudes toward trafficking in persons
expressed through the use of adverbs and
adjectives.

(Continued)
 Analyzing Data and Words—Guiding Principles ◾ 395

Table 12.23 (Continued ) Use Case 7: Trafficking in Persons Semantic Profile

Dimension Use Case Context

Semantic methods used Part of speech tagging, concept extraction,

categorization.

Design workflow Language workflow is followed.

Competencies needed Knowledge of trafficking from multiple

perspectives, the language of trafficking in a
diplomatic context, knowledge of extensional
and intentional language practices, and general
semantic methods.

Return on investment Potentially significant if the profile can be used to

identify and integrated all treatments of
trafficking across the organization. At this time,
that information is scattered and challenging to
find.

Qualitative risk reduction This is a highly sensitive and emotional topic.

Subjective interpretation is an inherent
challenge. In addition, the information on this
topic is hard to find—it is not a comfortable
topic for most to talk about or read about.

Feasibility without Not economically feasible across the

semantic tools organization without machine-based methods.
Manual approach is also at high risk for
subjective interpretation.

Table 12.24 Successes Achieved in the Sample Use Cases

Use Case
Number Success Factors

1 Topical characterization of content of diplomatic cables.

Discovery of intelligence content in the cables.

2 Use of Gottschalk–Gleser semantic profile to characterize political

discourse and media coverage.

3 Initial focus on linguistic registers and knowledge structures which

exposed the abstract and symbolic nature of mathematical language
and its relation to young students’ ability to learn concepts.

4 Development of a profile for discussing and exposing mistakes and

failures. This profile can be reused in other contexts in the future.

(Continued)
396 ◾ Analytics and Knowledge Management

Table 12.24 (Continued) Successes Achieved in the Sample Use Cases

Use Case
Number Success Factors

5 Both approaches produce high levels of type 1 and type 2 errors

which had heretofore not been investigated in depth.

6 Discovery of existing doctor-patient communication profiles by

experts in the field.

7 Having an authoritative reference model and nationally

recognized experts to work with in constructing and testing the
profile.

Fitting Tools to Methods and Sources

Organizations in the industrial and the knowledge economy should use every ana-
lytical method and every source available to achieve their goals. It is easier to select
and apply quantitative analytics than qualitative analytics, and there are more
quantitative analysis products on the market today. Qualitative analytics, though,
holds great promise—particularly for knowledge organizations and for knowledge
management researchers. Regardless of the type of organization, the business or
research focus, it is important to choose a tool that fits the methods and the sources
of evidence. The 27 questions can serve as a guide ensuring a good fit.
The challenge we face when selecting a tool is that not all of the commercially or
open-source tools described as “semantic” are constructed on top of such a foundation.
The technologies we need to do qualitative analysis should be built on a strong founda-
tional understanding of language—at least one language. They must be able to inter-
pret any content we give them at a primitive language level. They must also be able to
take further instructions from us on what to do with their understanding of the con-
tent. Knowledge scientists and business analysts must be able to instruct a computer to
extract and categorize the adjectives used in a corpus of text or speech, or to describe
the use of pronouns, or conditional verbs. These can be important analytical elements.
In this chapter, we distinguish between truly semantic technologies—which have
these capabilities—and stochastic technologies—which do not have these capabilities.
A stochastic application will apply quantitative or statistical methods to text with the
intent of determining frequencies or descriptive statistics of the use of words, or infer
categories from the patterns of usage it finds in the text. These types of technologies
may provide some level of information, but they are not comparable to the level of
qualitative analysis that is performed by humans based on human understanding.
While the market appears to be rich and robust, it is also volatile and difficult to
navigate. A review of the products on the market over the past ten years might show
that over a 3- year period close to 40% of the tools will have disappeared or have been
 Analyzing Data and Words—Guiding Principles ◾ 397

integrated into another product. The volatility is more often found on among lan-
guage and linguistics focused tools because they require a significant investment to
create and to sustain. These are also the tools that have the greatest value for enterprise
applications, so they are often purchased and integrated as a component of another
transactional or content management system. As consumers and producers of these
tools, we need to lobby for a more accurate portrayal of the functionality of the tools.

Reviewing Your Tool Choices

If you have addressed all of the key questions to this point, you’re ready to select a
tool or multiple tools to support your project. You should be prepared to develop
functional requirements, evaluate and test products before your purchase or acquire.
While acquiring a “free” or open source tool might seem attractive, it could be a
poor choice that suboptimizes your business or research goals. You should be pre-
pared to explain your choice based on evidence, not on marketing materials, a sales
promotion, or a conference presentation. You should be able to explain which tools
you considered and how you evaluated the tools. And you should consider a proof
of concept test as a condition of the procurement.

Reviewing Your Successes

The field of analytics and the field of knowledge management are emerging, even
though both can be traced back to the 1960s. It is important for everyone who is
working or experimenting in this area to share what they have learned, both suc-
cesses and failures. Did you achieve your goal? Did the tools behave as expected? Was
the solution affordable? What other costs did you incur? Did your project generate
new research projects? Did it change the way we think about a business issue? Were
you able to explore a topic that would not have been possible without these tools?
Table 12.24 describes how we would describe our successes if they were applied to the
use cases. We note that none of the success factors is related to a tool selection or to
the functionality of the tool. This is because the tool selection was the last decision.

Exposing Your Failures

It is equally important to share your failures and lessons learned. Where did your
solution fall short? How would you change the strategy if you were to reuse it in
the future? Were the results valid? Were they reliable? Based on what you learned,
would you not undertake the project? Were your expectations reasonable or unrea-
sonable? What was the most significant thing you learned? How would you explain
your failure to achieve your goals? Table 12.25 describes the lessons learned and
failures experienced in the use cases. We note that none of the lessons learned were
related to or a result of a tool selection or to the functionality of the tool. Again, this
is because the tool selection was the last decision.
398 ◾ Analytics and Knowledge Management

Table 12.25 Lessons Learned from the Sample Use Cases

Use Case
Number Lessons Learned

1 Late discovery that Wikileaks was credited with causation of Arab

Spring by only a few, but highly visible, media outlets. The more
accurate characterization was the increased use of social media
by those involved in the protests to chronicle the activities.

2 Difficulty of discovering and obtaining the political transcripts and

the variability of quality of the texts. Need for four distinct tools
to support he research.

3 Rather than generating registers of full textbooks and other

materials, we could have been more selective of portions of
textbooks. Having generated the linguistic register, we can now
adapt our future strategy.

4 We learned that the three months of communications were

representative of the most active periods of the year. A more
representative sample would have covered a full calendar year.

5 The extensive development of the country categorization profile

introduced some errors particularly across countries in a
region—where there were common naming practices.

6 Labor intensive effort to transcribe the videotaped interactions.

Next round of research we might test audio-to-text transcription
methods.

7 The size of the authoritative reports and the need to break them
into facet-focused chunks prior to doing any linguistic analysis or
extracting any knowledge structures.

Lessons Learned and Future Work

Working from the guiding questions, the use cases and other practical experience,
we offer 15 general lessons learned.

Lesson learned 1: Knowledge organizations have much to gain from wise and
considered use of the full range of analytical tools.
Lesson learned 2: Using analytics wisely means understanding the difference
between data and text analytics, good practice business intelligence models
and methods, and the nature and representation of information and knowl-
edge sources.
 Analyzing Data and Words—Guiding Principles ◾ 399

Lesson learned 3: Managing expectations is critical—promotional materials may

lead administrators to think they have found a silver bullet solution to quali-
tative analysis. There is a need to explain how the tools work, what they will
produce, and what resources are required to achieve those results.
Lesson learned 4: The availability of new smart tools does not override the need
for good analytical design. Fitting the right methods to the source data and
aligning with business or research goals will likely involve multiple tools and
methods. There are no silver bullet solutions on the analytics or the knowl-
edge management side. As the use cases suggest, the challenges and risks
associated with poor choices and poor designs are high. Wise design choices
can generate meaningful insights.
Lesson learned 5: Knowledge management methods such as knowledge model-
ing, knowledge elicitation, and knowledge representation are critical to the
wise use of analytical tools. The profession needs to be more deliberate in
their use with analytics.
Lesson learned 6: Validity, reliability, and trust risks arise when applying data
analytics tools to knowledge expressed in language. These risks are mitigated
where we approach data analytics with a deep understanding of language and
qualitative sources, and apply the appropriate analytical methods.
Lesson learned 7: Mixing methods and sources does not create a new analyti-
cal method and it certainly does not create a better analytical strategy. The
application of quantitative methods to qualitative data introduces a high risk
of error.
Lesson learned 8: Text analytics, business analytics and semantic analysis are not
WYSIWYG. Smart choices are made by researchers and decision makers who
understand methods and sources and who are willing to critically examine
the tools.
Lesson learned 9: The burden of bridging the gaps between business analytics
and knowledge management—to create the common ground—lies primarily
with the knowledge management discipline because these professionals have
a deeper understanding of knowledge, language, and qualitative methods.
Lesson learned 10: Academics and practitioners should work together to expand
the body of knowledge around language-based semantic applications.
Academic-business partnerships and academic-developer partnerships can go
a long way to growing a more robust and stable market. Most universities or
institutes are not set up to undertake this level of investigation, the best tools
are too expensive for academic researcher to use, and open source tools are
often unusable by anyone other than the developer.
Lesson learned 11: The tool developers and promoters should be encouraged to
describe the functionality of their products and services accurately, and con-
sumers must be willing and able to critique and provide constructive feed-
back to developers.
400 ◾ Analytics and Knowledge Management

Lesson learned 12: Text analytics and semantic analysis methods are not silver
bullets. They require investment to set up, configure, and sustain. They have
significant costs so most businesses get one chance to make the right choice.
Lesson learned 13: Research papers should describe in detail the tools used and
their capabilities. Peer reviewers should be willing to critique tool choices and
to question their use.
Lesson learned 14: The two most significant challenges for selecting a tool are
understanding the nature of the products that are available on the market
today, and working towards a stable and growing market for tools.
Lesson learned 15: The volatility of the market is related to the level of vendor
investment required to sustain these tools and the level of organizational invest-
ment required to maintain and operationalize them. As a result, the “quick-fix”
tools tend to dominate the market. A greater presence in the market, though,
does not signal a long product life span—the text analytics tools disappear from
the commercial market at the same rate as the language-based applications.
Language-based products may have a longer life span where they are integrated
into a larger enterprise application (e.g., SAP, SAS). The effect for the consumer
is the same, though—they disappear from the open market. Tools that support
academic research—computational linguistic tools—tend to have a longer life
span. However, these tools are generally not business or designer-friendly.

Conclusions
We know from experience that we can improve the effectiveness of data and text
analytics by leveraging knowledge management methods such as knowledge model-
ing, knowledge representation, and knowledge engineering. The opportunities and
rewards of applying analytics to knowledge management challenges hold equal oppor-
tunities. The opportunities for knowledge management professionals lie primarily in
the transformation of qualitative methods to machine-based methods. The challenge,
though, is the amount of design, thinking, and investment required to achieve those
results. What we have learned over the past 45 years is that this is possible if we have
dedicated academic and development resources to support the transformation. The
research questions we can pose and explore have increased exponentially. These ques-
tions would have been too manually intensive or too time consuming to investigate
in the past. Business decisions that have had to rely on experience and “gut instincts”
can now be tested with machine-based qualitative solutions.
We need to learn to think more expansively and creatively in defining our
research agendas and setting our business goals. Creativity and discovery begins
with the first strategic dimension—setting expectations and the framework for
analysis and solution. It means modeling knowledge and thought processes, apply-
ing appreciative inquiry process at every step of the way, expecting to do discov-
ery at a micro level with macro level impacts. It means having the opportunity to
 Analyzing Data and Words—Guiding Principles ◾ 401

increase our understanding of human knowledge and remapping knowledge, and

achieving greater exposure for tacit knowledge. Knowledge management practitio-
ners also need to deepen the language-focused components of the discipline. This
means collaborating with developers of analytics tools, and with bridging the gaps
that exist across disciplines.

References
Abbasi, A. A., & Younis, M. (2007). A survey on clustering algorithms for wireless sensor
networks. Computer Communications, 30(14), 2826–2841.
Abilhoa, W. D., & De Castro, L. N. (2014). A keyword extraction method from Twitter
messages represented as graphs. Applied Mathematics and Computation, 240, 308–325.
Adcock, R. (2001). Measurement validity: A shared standard for qualitative and quantita-
tive research. American Political Science Review, 95(3), 529–546.
Anwar, W., Wang, X., & Wang, X. L. (2006). A survey of automatic Urdu language processing. In
Machine learning and cybernetics, 2006 international conference on (pp. 4489–4494). IEEE.
Arya, S., Mount, D. M., Netanyahu, N. S., Silverman, R., & Wu, A. Y. (1998). An optimal
algorithm for approximate nearest neighbor searching fixed dimensions. Journal of the
ACM (JACM), 45(6), 891–923.
Atkinson, P. (2005). Qualitative research—Unity and diversity. In Forum qualitative sozial-
forschung/forum: Qualitative social research (Vol. 6, No. 3). London: Routledge.
Babbie, E. R. (1989). The practice of social research. Belmont, CA: Wadsworth Publishing Company.
Ballard, B. W., & Biermann, A. W. (1979, January). Programming in natural language: “NLC” as a
prototype. In Proceedings of the 1979 annual conference (pp. 228–237). New York, NY: ACM.
Bar-Hillel, Y. (1966). Language and information; selected essays on their theory and applica-
tion. Reading, MA: Addison-Wesley.
Barton, D., & Hamilton, M. (2012). Local literacies: Reading and writing in one community.
London: Routledge.
Bedford, D., & Platt, C. (2013, September 16). Mathematical languages—Barriers to knowl-
edge transfer and consumption. Scientific Information Policies in the Digital Age:
Enabling Factors and Barriers to Knowledge Sharing and Transfer, Aula Marconi,
Consiglio Nazionale delle Ricerche, Rome, Italy.
Bedford, D. A. D., Bekbalaeva, J., & Ballard, K. (2017, November 27–31). Global human
trafficking seen through the lens of semantics and text analytics. American Society for
Information Science and Technology Annual Conference, Crystal City, VA.
Bedford, D. A. D., Turner, J., Norton, T., Sabatiuk, L., & Nassery, H. (2017, November
27–31). Knowledge translation in health sciences. American Society for Information
Science and Technology Annual Conference, Crystal City, VA.
Bedford, D. A. D. (2012a). Enhancing the precision of geographical tagging—Embedding gazetteers
in semantic analysis technologies. Poster session presented at the TKE 2012, Madrid, Spain.
Bedford, D. A. D. (2012b). Semantic analysis of the political discourse in the presidential
and congressional campaigns of 2012. Paper presented at the Text Analytics World
Conference, San Francisco, CA.
Bedford, D. A. D., & Neville, L. (2016, September). Knowledge sharing and valuation in
beekeeping communities. In Proceedings of the international conference on intellectual
capital knowledge management and organizational learning. Ithaca, NY: Ithaca College.
402 ◾ Analytics and Knowledge Management

Bejou, D., Wray, B., & Ingram, T. N. (1996). Determinants of relationship quality: An arti-
ficial neural network analysis. Journal of Business Research, 36(2), 137–143. Oxford:
Oxford University Press
Bengio, Y. (2009). Learning deep architectures for AI. Foundations and Trends® in Machine
Learning, 2(1), 1–127.
Berman, R. A. (2014). Linguistic perspectives on writing development. In B. Arfe,
J. Dockrell, & V. Berninger (Eds.), Writing development in children with hearing
loss, dyslexia or oral language problems: Implications for assessment and instruction
(pp. 176–186). Oxford: Oxford University Press.
Berman, R., & Nir, B. (2009). Cognitive and linguistic factors in evaluating text quality:
Global versus local. In V. Evans & S. Pourcel (Eds.), New directions in cognitive linguistics
(pp. 421–440). Amsterdam, the Netherlands: John Benjamins.
Berman, R., & Verhoeven, L. (2002). Cross-linguistic perspectives on the development of
text-production abilities: Speech and writing. Written Language and Literacy, 5(1), 1–43.
Biermann, A. W. (1981). Natural language programming. In A. Biermann & G. Guiho
(Eds.), Computer program synthesis methodologies (pp. 335–368). Amsterdam, the
Netherlands: Springer.
Biermann, A. W., Ballard, B. W., & Sigmon, A. H. (1983). An experimental study of natural
language programming. International Journal of Man-machine Studies, 18(1), 71–87.
Bollinger, A. S., & Smith, R. D. (2001). Managing organizational knowledge as a strategic
asset. Journal of Knowledge Management, 5(1), 8–18.
Brandow, R., Mitze, K., & Rau, L. F. (1995). Automatic condensation of electronic publica-
tions by sentence selection. Information Processing and Management, 31(5), 675–685.
Carletta, J. (1996). Assessing agreement on classification tasks: The kappa statistic.
Computational Linguistics, 22(2), 249–254.
Carmines, E. G., & Zeller, R. A. (1979). Reliability and validity assessment (Vol. 17).
Newbury Park, CA: Sage Publications.
Chilton, P. (2004). Analysing political discourse: Theory and practice. London: Routledge.
Chowdhury, G. G. (2003). Natural language processing. Annual Review of Information
Science and Technology, 37(1), 51–89.
Church, K. W., & Hanks, P. (1990). Word association norms, mutual information, and
lexicography. Computational Linguistics, 16(1), 22–29.
Church, K. W., & Mercer, R. L. (1993). Introduction to the special issue on computational
linguistics using large corpora. Computational Linguistics, 19(1), 1–24.
Cope, B., & Kalantzis, M. (Eds.). (2000). Multiliteracies: Literacy learning and the design of
social futures. London: Psychology Press.
Croft, D., Coupland, S., Shell, J., & Brown, S. (2013). A fast and efficient semantic short
text similarity metric. Computational Intelligence (UKCI), 2013 13th UK Workshop on
(pp. 221–227). IEEE.
Dahl, V., & Saint-Dizier, P. (1985). Natural language understanding and logic programming.
New York, NY: Elsevier.
Debortoli, S., Müller, O., Junglas, I. A., & vom Brocke, J. (2016). Text mining for informa-
tion systems researchers: An annotated topic modeling tutorial. CAIS, 39, 7.
Deerwester, S., Dumais, S. T., Furnas, G. W., Landauer, T. K., & Harshman, R. (1990). Indexing
by latent semantic analysis. Journal of the American Society for Information Science, 41(6), 391.
Dijkstra, E. W. (1979). On the foolishness of “natural language programming”.
In F. L. Bauer, E. W. Dijkstra, S. L. Gerhart, & D. Gries (Eds.), Program construction
(pp. 51–53). Berlin: Springer.
 Analyzing Data and Words—Guiding Principles ◾ 403

Edmundson, H. P. (1969). New methods in automatic extracting. Journal of the ACM

(JACM), 16(2), 264–285.
Eggins, S. (2004). Introduction to systemic functional linguistics. London: A&C Black.
Eggins, S., & Slade, D. (2005). Analysing casual conversation. London: Equinox Publishing.
Eriksson, P., & Kovalainen, A. (2015). Qualitative methods in business research: A practical
guide to social research. London: Sage Publications.
Fairclough, N. (2003). Analysing discourse: Textual analysis for social research. London:
Psychology Press.
Fairclough, N. (2013). Critical discourse analysis: The critical study of language. London: Routledge.
Ferguson, C. A. (1983). Sports announcer talk: Syntactic aspects of register variation.
Language in Society, 12(2), 153–172.
Fernández, A., Gómez, Á., Lecumberry, F., Pardo, Á., & Ramírez, I. (2015). Pattern recog-
nition in Latin America in the “Big Data” era. Pattern Recognition, 48(4), 1185–1196.
Finegan, E., & Biber, D. (1994). Register and social dialect variation: An integrated
approach. In D. Biber & E. Finegan (Eds.), Sociolinguistic perspectives on register
(pp. 315–347). Oxford: Oxford University Press.
Gee, J. (2015). Social linguistics and literacies: Ideology in discourses. New York, NY: Routledge.
Goddard, C. (2011). Semantic analysis: A practical introduction. Oxford: Oxford University Press.
Guion, R. M. (1980). On Trinitarian doctrines of validity. Professional Psychology, 11(3), 385.
Halliday, M. A. K., McIntosh, A., & Strevens, P. (1964). The linguistic sciences and language
teaching. London: Longmans.
Halpern, M. (1966). Foundations of the case for natural-language programming. In
Proceedings of the November 7–10, 1966, all joint computer conference (pp. 639–649).
San Franscisco, CA: ACM.
Han, J., Pei, J., & Kamber, M. (2011). Data mining: Concepts and techniques. San Diego, CA: Elsevier.
Hand, D. J., Mannila, H., & Smyth, P. (2001). Principles of data mining. Cambridge, MA: MIT Press.
Hatzivassiloglou, V., & McKeown, K. R. (1997, July). Predicting the semantic orientation
of adjectives. In Proceedings of the eighth conference on European chapter of the asso-
ciation for computational linguistics (pp. 174–181). Morristown, NJ: Association for
Computational Linguistics.
Hearst, M. A. (1992, August). Automatic acquisition of hyponyms from large text corpora.
In Proceedings of the 14th conference on computational linguistics (pp. 539–545, Vol. 2).
Nantes: Association for Computational Linguistics.
Heidorn, G. E. (1971). Natural language inputs to a simulation programming system: An
introduction. Monterey, CA: Naval Postgraduate School.
Henzl, V. M. (1973). Linguistic register of foreign language instruction. Language Learning,
23(2), 207–222.
Hindle, D., & Rooth, M. (1993). Structural ambiguity and lexical relations. Computational
Linguistics, 19(1), 103–120.
Hofmann, T. (1999). Probabilistic latent semantic indexing. In Proceedings of the 22nd
annual international ACM SIGIR conference on research and development in informa-
tion retrieval (pp. 50–57). Berkeley, CA: ACM.
Jurafsky, D. (1996). A probabilistic model of lexical and syntactic access and disambigua-
tion. Cognitive Science, 20(2), 137–194.
Kaplan, D., & Berman, R. (2015). Developing linguistic flexibility across the school years.
First Language, 35(1), 27–53.
King, G., Keohane, R. O., & Verba, S. (1994). Designing social inquiry: Scientific inference
in qualitative research. Princeton, NJ: Princeton University Press.
404 ◾ Analytics and Knowledge Management

Kirk, J., & Miller, M. L. (1986). Reliability and validity in qualitative research. Beverly Hills,
CA: Sage Publications.
Kress, G. (2009). Multimodality: A social semiotic approach to contemporary communication.
London: Routledge.
Kruger, D. J. (2003). Integrating quantitative and qualitative methods in community
research. The Community Psychologist, 36(2), 18–19.
Lehnert, W., Soderland, S., Aronow, D., Feng, F., & Shmueli, A. (1995). Inductive text clas-
sification for medical applications. Journal of Experimental and Theoretical Artificial
Intelligence, 7(1), 49–80.
Liddy, E. D. (2001). Natural language processing. Syracuse, NY: Syracuse University.
Loper, E., & Bird, S. (2002, July). NLTK: The natural language toolkit. In Proceedings
of the ACL-02 workshop on effective tools and methodologies for teaching natural lan-
guage processing and computational linguistics (pp. 63–70, Vol. 1). Philadelphia, PA:
Association for Computational Linguistics.
Luhn, H. P. (1958a). A business intelligence system. IBM Journal of Research and
Development, 2(4), 314–319.
Luhn, H. P. (1958b). The automatic creation of literature abstracts. IBM Journal of Research
and Development, 2(2), 159–165.
Maarek, Y. S., Berry, D. M., & Kaiser, G. E. (1991). An information retrieval approach
for automatically constructing software libraries. IEEE Transactions on Software
Engineering, 17(8), 800–813.
Manning, C. D., & Schütze, H. (1999). Foundations of statistical natural language processing
(Vol. 999). Cambridge, MA: MIT Press.
Marcu, D. (1997, July). The rhetorical parsing of natural language texts. In Proceedings of
the 35th annual meeting of the association for computational linguistics and eighth confer-
ence of the European chapter of the association for computational linguistics (pp. 96–103).
Madrid: Association for Computational Linguistics.
Marcus, M. P., Marcinkiewicz, M. A., & Santorini, B. (1993). Building a large annotated
corpus of english: The penn treebank. Computational Linguistics, 19(2), 313–330.
Berlin, Heidelberg: Springer.
Mihalcea, R., Liu, H., & Lieberman, H. (2006, January). NLP (natural language process-
ing) for NLP (natural language programming). In A. Belbukh (Ed.), International
Conference on intelligent text processing and computational linguistics (pp. 319–330).
Berlin, Heidelberg: Springer.
Miner, G. (2012). Practical text mining and statistical analysis for non-structured text data
applications. Amsterdam, the Netherlands: Academic Press.
Mitkov, R. (Ed.). (2005). The Oxford handbook of computational linguistics. Oxford: Oxford
University Press.
Neumann, S. (2014). Cross-linguistic register studies: Theoretical and methodological con-
siderations. Languages in Contrast, 14(1), 35–57.
Nir, B., & Berman, R. (2010). Parts of speech as constructions: The case of Hebrew
“adverbs”. Constructions and Frames, 2(2), 242–274.
Pereira, F., Tishby, N., & Lee, L. (1993, June). Distributional clustering of English words.
In Proceedings of the 31st annual meeting on association for computational linguistics
(pp. 183–190). Montreal: Association for Computational Linguistics.
Pustejovsky, J. (1991). The generative lexicon. Computational Linguistics, 17(4), 409–441.
Ravid, D., & Berman, R. (2009). Developing linguistic register across text types: The case
of Modern Hebrew. Pragmatics and Cognition, 17(1), 108–145.
 Analyzing Data and Words—Guiding Principles ◾ 405

Ravid, D., & Berman, R. A. (2010). Developing noun phrase complexity at school age:
A text-embedded cross-linguistic analysis. First Language, 30(1), 3–26.
Reason, P., & Bradbury, H. (Eds.). (2001). Handbook of action research: Participative inquiry
and practice. London: Sage Publications.
Salton, G. (1970). Automatic text analysis. Science, 168(3929), 335–343.
Salton, G. (1986). Another look at automatic text-retrieval systems. Communications of the
ACM, 29(7), 648–656.
Schiffman, H. (1997). The study of language attitudes. Philadelphia, PA: University of
Pennsylvania.
Seale, C. (1999). Quality in qualitative research. Qualitative Inquiry, 5(4), 465–478.
Shieber, S. M., Schabes, Y., & Pereira, F. C. (1995). Principles and implementation of
deductive parsing. The Journal of Logic Programming, 24(1), 3–36.
Shmueli-Scheuer, M., Roitman, H., Carmel, D., Mass, Y., & Konopnicki, D. (2010).
Extracting user profiles from large scale data. In Proceedings of the 2010 workshop on
massive data analytics on the cloud (p. 4). New York, NY: ACM.
Spigelman, J. J. (1999). Statutory interpretation: Identifying the linguistic register.
Newcastle Law Review, 4, 1.
Sproat, R., Black, A. W., Chen, S., Kumar, S., Ostendorf, M., & Richards, C. (2001).
Normalization of non-standard words. Computer Speech and Language, 15(3),
287–333.
Spyns, P. (1996). Natural language processing. Methods of Information in Medicine, 35(4),
285–301.
Srinivas, M., & Patnaik, L. M. (1994). Genetic algorithms: A survey. Computer, 27(6),
17–26.
Tesch, R. (2013). Qualitative research: Analysis types and software. London: Routledge.
Teufel, S., & Moens, M. (2002). Summarizing scientific articles: Experiments with rel-
evance and rhetorical status. Computational Linguistics, 28(4), 409–445.
Tischer, S. (2009). U.S. Patent No. 7,483,832. Washington, DC: U.S. Patent and Trademark
Office.
Tomita, M. (2013). Efficient parsing for natural language: A fast algorithm for practical systems
(Vol. 8). New York, NY: Springer Science and Business Media.
Ure, J. (1982). Introduction: Approaches to the study of register range. International Journal
of the Sociology of Language, 1982(35), 5–24.
Van Gijsel, S., Geeraerts, D., & Speelman, D. (2004). A functional analysis of the linguis-
tic variation in Flemish spoken commercials. In G. Purnelle, C. Fairon, & A. Dister
(Eds.), Le poids des mots. Proceedings of the 7th international conference on the statistical
analyses of textual data (pp. 1136–1144). Louvain-la-neuve: Presses Universitaires de
Louvain.
Weizenbaum, J. (1966). ELIZA—A computer program for the study of natural language
communication between man and machine. Communications of the ACM, 9(1),
36–45.
Witten, I. H., Frank, E., Hall, M. A., & Pal, C. J. (2016). Data mining: Practical machine
learning tools and techniques. Burlington, VT: Morgan Kaufmann.
Yarowsky, D. (1995, June). Unsupervised word sense disambiguation rivaling supervised
methods. In Proceedings of the 33rd annual meeting on association for computa-
tional linguistics (pp. 189–196). Stroudsburg, PA: Association for Computational
Linguistics.
Chapter 13

Data Analytics for Cyber

Threat Intelligence
Hongmei Chi, Angela R. Martin, and Carol Y. Scarlett

Contents
Introduction ..................................................................................................... 408
Cyber Threat Intelligence ................................................................................. 409
Related Work in Cyber Threat Intelligence ...................................................410
Computational Methods in Cyber Treat Intelligence..................................... 411
Challenges in Cyber Threat Intelligence ........................................................413
Prevalence of Social Media Analysis in Cyber Threat Intelligence .................. 415
Benefits of Social Media Analysis in Cyber Threat Intelligence ......................416
Establishing Motivations in Cyber Threat Intelligence through Social
Media Analysis.............................................................................................. 417
Role of Behavioral and Predicative Analysis in Cyber Threat Intelligence ......418
Text Analysis Tools ........................................................................................... 420
Linguistic Inquiry Word Count ....................................................................421
Sentiment Analysis ...................................................................................... 423
SentiStrength ................................................................................................424
Case Study Using Linguistic Inquiry Word Count ........................................425
Conclusions ......................................................................................................427
References .........................................................................................................427

407
408 ◾ Analytics and Knowledge Management

Introduction
Cybersecurity nowadays is a national priority due to the widespread application of
information technology and the growing incidence of cybercrime. Data science and
big data analysis is gaining popularity due to widespread utility in many subfields
of cybersecurity, such as Intrusion detection system (IDS), social networks, insider
threat, and wireless body network. The use of advanced analytic techniques, com-
putational methods, and traditional components that have become representative
of “data science” has been at the center of cybersecurity solutions focused on iden-
tification and prevention of cyberattacks.
Most companies are capable of hardening their external defenses using various
information security methods available. However, companies often fail in protecting
against attacks that originate from within the company. These are known as insider
threats that are difficult to guard against, because employees are given a certain level
of access to internal company networks that normally bypass the external security
measures put in place to protect the company. One way to identify insider threats is
through the use of predictive analysis monitoring social networks. Social network-
ing is a form of communication over virtual spaces (Castells, 2007). Monitoring
such networks requires tools that can readily analyze linguistic patterns.
Analyzing the huge amount of content in social networks needs linguistic tools
such as linguistic inquiry word count (LIWC), Stanford’s Core NLP Suite, or
SentiStrength. Natural language processing is the ability of computers to under-
stand what users are writing using frequently used human writing patterns. This
means the computer must be capable of identifying various dialects, slang, and
even properly identify homonyms to better interpret human language. Most NLP
algorithms are based on statistical machine learning techniques, which use statisti-
cal inferences. These inferences come from rules gained from analyzing large sets
of documents known as corpora which contain the correct values that need to be
studied (Chi et al., 2016). An example of using NLP is the auto selection in the
Google search bar. As the user types a query into the search bar, Google’s search
algorithm considers the words that the user has already typed and compares them
to the words most often typed by other users. This allows Google to provide users
with options and the ability to auto-complete the remaining statement based on
these similar searches. Such tools could aid in detecting insider threats associated
through identifying conversations that show some agenda or malice.
In this chapter, we focus on the potential use of computational methods in
intelligence work, insider security threats, and mobile health. The standard bearer
for mobile computing is the smartphone, which connects people anytime and
anywhere. Smartphones also store large amounts of personal information and run
applications that may legitimately, inadvertently, or maliciously manipulate per-
sonal information. The relatively weak security models for smartphone applica-
tions, coupled with ineffective security verification and testing practices, have made
smartphones an ideal target for security attacks. Advancing the science, technology
 Data Analytics for Cyber Threat Intelligence ◾ 409

Analysis
Collection Processing Decision
(machine Dissemination
(social media) (NLP) support
learning)

Figure 13.1 Cyber intelligence processing.

and practices for securing mobile computing are essential to support the inevitable
use of mobile computing in areas with even requirements for privacy, security, and
resistance to tampering (Figure 13.1).

Cyber Threat Intelligence

Cyber threat intelligence (CTI) has become a primary focus in the intelligence
community as cyber threats invade almost every branch of national security. On
February 23, 2000, the Central Intelligence Agency (CIA) briefed the Senate Select
Committee on Intelligence that foreign cyber threat was one of the key challenges
to the nation acknowledging the issue as a developing form of warfare for weaker
nations unable to compete with the United States in terms of raw power. In this
briefing, the CIA emphasized that both China and Russia saw cyber warfare as
a viable tool of war. The briefing also pointed out that cyber capabilities in the
hands of terrorists posed an uncontrollable threat as non-state actors are devoid of
normal deterrents that would prevent powerful nations from conducting cyberat-
tacks (CIA, 2000). In 2002, the Federal Information Security Management Act
(FISMA), was passed into federal law acknowledging cybersecurity as inherent
to the nation’s security and mandating agency wide program development (U.S.
Congress, H.R. 3844, 2002).
Since then, the United States has experienced spectacular, costly, and embar-
rassing cyberattacks, the exposure of which has had major implications for U.S.
security. Titan Rain is the code name for Chinese-based cyberattacks against U.S.
government computers that began in 2003 and continued for over several years
stealing terabytes of government data (Thornburgh, 2005). The Department of
Homeland Security (DHS) computers were hacked over a period of three months
in 2006 (Nakashima and Krebs, 2007). The Secretary of Defense was hacked to
exploit the Pentagon in 2007 (Sevastopulo, 2007). The National Security Agency
(NSA) revealed in a high-level report that the private emails of top officials in
Obama’s administration were hacked in 2010 (Windrem, 2015). Iran, in 2011,
410 ◾ Analytics and Knowledge Management

claimed to have hacked a U.S. drone (Rawnsley, 2011). Although the United States
never publicly acknowledged the hack, Obama did acknowledge the capture of
the drone (CNN Wire Staff, 2011). In response to these events, the DOD Defense
Science Board (2013) released a report detailing their assessment of the cyberse-
curity threat to national security with the recommendation that the U.S. govern-
ment take cyber threat seriously and rethink the way they operate (2013). Shortly
thereafter, in March of 2014, the Office of Personnel Management (OPM) systems
and two government contractors were hacked affecting over 4.2 million employ-
ees (Koerner, 2016). The Senate, in response to this attack, passed the Federal
Information Security Management Act of 2014 which redirected the security over-
sight of the OPM systems to the Department of Homeland Security and established
a Federal Information Security Incident Center (U.S. Congress, S-2521, 2014).
Despite the government’s best efforts to develop and improve the nation’s cyber-
security posture, the implementation plan did not happen fast enough. In 2015, the
Pentagon’s joint staff unclassified email system was taken down for two weeks due
to an attack suspected to originate from Russia (Bennett, 2015). The same OPM
hacker, in a similar attack, stole personal information from 21.5 million people all
related to government service (Davis, 2015). Events such as these brought to the
forefront the scale of vulnerability within the intelligence community. In March of
2017, Wikileaks published a treasure trove of professed CIA hacking capabilities,
the implications of which have yet to be fully understood (Lapowsky and Newman,
2017). In April of 2017, an NSA hacking tool called EternalBlue was stolen and
released online (Larson, 2017). This tool was then used in the WannaCry hack in
May 2017 that reportedly affected 99 countries (BBC Europe, 2017). These events
showed that even the most secure organizations are vulnerable and the threat is
moving at a faster rate than CTI. The government quickly realized that advanced,
efficient, and rapid solutions are needed to secure intelligence information used to
safeguard the nation.

Related Work in Cyber Threat Intelligence

The systemization of the CTI field is in the development stage but preliminary
works to provide structure and organization are evident (Lee, 2014). A patented
research study, completed in 2002, systematized CTI into a series of standardized
steps that included data collection, processing, analysis, record creation, knowledge
base creation, product creation, and client information collection (Edwards et al.,
2002). Cloppert, from the SANS institute, categorized indicators of attacks into
three categories: atomic (things that cannot be broken down like IP addresses and
emails), computed (things that can be broken down like code and hash tags) and
behavioral (attack patterns) (2009). Lockheed Martin formalized what is known
as the Cyber Kill Chain to describe the stages of attack that include reconnais-
sance, weaponization, delivery, exploitation, installation, command and control,
and actions on objectives. (Hutchins et al., 2011).
 Data Analytics for Cyber Threat Intelligence ◾ 411

Since then, there have been several related works and attempts to categorize
indicators of intrusion like the Cyber Observable Expression (CybOX), the Open
Indicators of Compromise (OpenIOC), and the Incident Object Description
Exchange Format (IODEF) frameworks (Gragido, 2012). Around the same time,
the Cyber Intelligence Sharing and Protection Act was passed which allows for
government to share intelligence with private corporations for the purpose of secur-
ing networks and systems (112 Congress, 2012). SANS provides an in-depth white
paper on some major CTI works titled “Tools and Standards for Cyber Threat
Intelligence Projects” (Farham, 2017). Shortly after these developments, the cyber-
security and communications section under the Department of Homeland Security
led a collaborative environment to establish a structured language for CTI sharing
titled Structured Threat Information eXpression (STIX) which uses the Trusted
Automated eXchange of Indicator Information (TAXII™) to exchange CTI infor-
mation between the government and private sectors. The structure consists of eight
core constructs (Barnum, 2014). The eight core constructs of STIX are campaign,
course of action, exploit target, incident, indicator, observable, threat actor, and
tactics, techniques, and procedures (TTP). Many of the earlier frameworks were
incorporated into STIX and the effort is now being led by the OASIS Cyber Threat
Intelligence Technical Committee (OASIS, 2017).

Computational Methods in Cyber Treat Intelligence

CTI has developed at amazing speed some impressive analytical visualization tools
to meet the growing demands for cyber threat indication. Table 13.1 demonstrates
the wealth of tools available to conduct CTI. Most of these tools have the primary
function of assisting the CTI analyst with collecting, categorizing, organizing, and
visualizing data quickly. Each tool may have its own special utility that makes it
stand out from the rest but there are some basic functions or tools that most of
these platforms should have to be competitive (Department of Justice, 2007). Some
common functions found in these platforms include data mining tools like word
cloud, link analysis, geographic information system (GIS) capabilities, and statisti-
cal visualization tools like charts and plots.
Ultimately, the success of these platforms will be defined by their utility to the
security analyst. Word cloud is a quantitative tool that provides visual displays of
frequently used words. Word clouds are useful to the security analyst in narrow-
ing the scope of study by allowing the security analyst to select reports that have
words of interest. They also help the security analyst understand the quantity of
reports they are dealing with. Word clouds can be useful when needing to separate
relevant nouns or action words from nonrelevant ones and can assist in finding a
word in a report like finding a needle in a haystack (this is usually a person). Link
analysis helps the security analyst establish associations particularly as they relate
to discovering central nodes in a network of interest or outliers that do not belong;
an example is Thycotic’s use of link analysis to recognize unauthorized access
412 ◾ Analytics and Knowledge Management

Table 13.1 Computational Platforms for Cyber Threat Intelligence

Brand Name Use Web Address

Anomali Threat intelligence platform https://www.anomali.com/

ArcSight Security intelligence http://www.ndm.net/

platform arcsight/

Cyber4Sight Predictive threat https://sso.cyber4sight.

intelligence platform com/idp/SSO.saml2

Haystack Constellation analytics https://haystax.com/

security platform

Maltego Open source data mining https://www.paterva.com/

tool web7/index.php

Metasploit Penetration testing https://www.metasploit.

software com/

Qualys Cloud based security https://www.qualys.com/

management tool

Palantir Big data analysis tool https://www.palantir.com/

Recorded Threat intelligence https://www.

future platform recordedfuture.com/

Securonix Security analytics platform https://www.securonix.

com/

Spectorsoft User behavior monitoring http://www.veriato.com/

software

Splunk Operational intelligence https://www.splunk.com/

platform

Open stack Open source cloud https://www.openstack.

computing software org/

Tableau Big data visualization tool https://www.tableau.com/

Threat Security operations and https://www.

connect analysis platform threatconnect.com/

Threat Threat intelligence https://www.threatq.com/

quotient platform
 Data Analytics for Cyber Threat Intelligence ◾ 413

(Thycotic, 2017). GIS capabilities help security analysts visualize attack patterns in
terms of geography. Geography plays a strong role in cyber activity as the actions of
a hacker correlate with the hacker’s environment even if the hacker operates in the
digital world (Bronk, 2016). An example of this is the MalwareTech botnet tracker
that uses GIS capabilities to track and visualize botnets throughout the globe
(MalwareInt, 2017). The Department of Defense realizes the relevancy of geogra-
phy to cyber activity demonstrated by the strategic objective to build and maintain
strong global alliances to combat cyber threats (The Department of Defense Cyber
Strategy, 2015). GIS capabilities can also assist with predicting future attack pat-
terns through the visualization of previous patterns. Data visualization tools help
conduct comparative or trend analysis such as heat signatures for identifying high
activity concentrations or graphs to identify escalation of events.
Regardless of the analytical capability, security analysts require tools that can
talk and communicate with each other for quick exchange of data sharing. Once
a threat is identified, time is of the essence which means the tool should facilitate
not only thorough and in-depth analysis but also be able to share information with
other security analysts quickly. Thus, the data should flow fluidly from one system
to another to give others adequate time to respond and secure their systems. As
demonstrated in Table 13.1, CTI tools are developed from a variety of sources in
response to the competitive and lucrative field but are not necessarily able to cross-
talk with other intelligence tools. The more the tool can communicate with the
other tools out there, the more utility it has. The usability of the tool will also be a
factor, as the security analyst is not necessarily going to be a subject matter expert
in computer science even though the intelligence community is certainly trying to
change this. Until then, the tool should be relatively easy for a security analyst to
use, as the most common security analyst tasks include the ability to read, inter-
pret, and apply that data to world events. As of right now, the more user friendly
the CTI tool, the more it will be used (Tableau, 2017).

Challenges in Cyber Threat Intelligence

The challenges in CTI are for the most part the same as any intelligence field. They
revolve around the ability to pull, store, find, and share reliable data. SANS institute
completed a study on CTI to establish a baseline for CTI. The results of the study
demonstrated that a demand exists to enable the aggregation of information from
all available sources, visualize that data in an expedient and comprehensible man-
ner, and verify the reliability of the data via multiple reporting sources (Shackleford,
2015). This is especially true in the intelligence community, as tools that pull from
unclassified source data, like social media sites, do not directly feed into classified
intelligence channels so the transfer of data is particularly challenging.
The ability to selectively collect and filter relevant information from the wealth
of data out there presents another challenge. Konkel reports that even though the
intelligence community is one of the largest collectors of Big Data, the information
414 ◾ Analytics and Knowledge Management

is fragmented and assembling the information is the real challenge (2014). CTI
currently relies on the collection of data to find signatures and anomalies to iden-
tify hackers but these techniques have overwhelmed security analysts. A research
study from Enterprise Strategy Group found that 74% of security operations teams
reported that security events and alerts are ignored due to volumes of data and not
enough staff (Davis, 2016). With so much data being collected, expiration dates on
the information become an issue as the data expires faster than the security analyst
can come to conclusions and act on those conclusions (Dickson, 2016). Expiration
dates are also an issue because, as one Big Data platform pointed out, threat analy-
sis is only released to the public after the fact to avoid tipping off the hacker that
someone is tracking the hack (SecureWorks, 2017).
One of the main attractions to cybercrime is the anonymity of the act. Experts
in the field are frustrated by the fact that origins of attack are difficult to trace
(J.G., 2014). Although the CTI professional can remove malware from a system,
the malware itself is not eradicated. Once the malware is identified and extracted,
a hacker can still change small portions of the code and then use the same malware
again. One piece of malware derived from one person can spread to thousands of
computers and then be manipulated in seconds to spread again (Koerner, 2016).
This highlights the futility of trying to track down a hacker from the trace, sig-
nature, design, or timestamp. One alternative response would be to identify the
source from its point of origin rather than from a trace. This is where monitoring
social media in conjunction with IT systems has the potential to lessen the time it
takes to identify a threat.
Identity management is another challenge in the intelligence field as the inter-
net offers the element of anonymity. For example, one report on social media sites
estimates that approximately 31 million Facebook user accounts are fake (Hayes,
2016). Regardless of how a tool tracks an identity, determining identities will
always be a challenge when dealing with the digital world as the slightest mistake
in reporting can change the identity of a person. Biometric tools have this chal-
lenge despite the capability of establishing unique identities, such as fingerprints.
The data that is assigned to the fingerprint still must be tagged. Should a finger-
print have similar biometric data to another individual, this information can eas-
ily become scrambled; many fingerprint readers use only selected points along the
fingerprint ridges for identity, which results in up to 3% misidentification or over-
lapping fingerprints between individuals. The intelligence world is thus turning to
machine learning and predictive analysis to remedy these issues.
The solution to these challenges is an obvious one: Track the hacker by develop-
ing easy to use tools that can collect data from a variety of independent sources in a
selective manner. These tools should be able to communicate with other tools in an
easy and timely fashion. Although the solution maybe is obvious, the implementa-
tion of the solution is not. Further development of machine learning programs is
required to selectively pull relevant information. This paper views cyber intelligence
as the computational ability to identify what information to collect to successfully
 Data Analytics for Cyber Threat Intelligence ◾ 415

identify the threat without collecting irrelevant data. With the advancement of
technology and the speed in which the intelligence world is moving towards cyber-
security, the ability to develop these types of computational methods is what will
define one Big Data analytic tool over another. This is also where algorithms of Big
Data have a place in cyber threat intelligence.

Prevalence of Social Media Analysis

in Cyber Threat Intelligence
Social media analysis has its place in the field of CTI in multiple ways, but for the
future of CTI, the attraction is primarily due to its predictive nature, because it has
the potential to establish more than just an identity. It can also establish moods,
tones, and feelings prior to acts. Text analysis through social media has been useful
in other areas of intelligence to establish security levels in an environment and to
detect when that level is changing. Establishing security levels is important when
one is concerned about the safety of those who operate in that space, be they gov-
ernment officials, military, law enforcement, intelligence professionals, or civilians.
Text analysis through social media provides insight into the status of populations,
such as levels of satisfaction using quantitative and qualitative analytics. The sig-
nificance of this is that social media analysis has the potential to discover the source
of a cyberattack, prior to the attack taking place, because of its ability to establish
human motivations and moods.
Data analytics, as it relates to social media, appears to be an underdeveloped
and underutilized source in the CTI field. In a threat intelligence capability brief-
ing, establishing motivations is listed under strategic threat intelligence but when
looking at the sources available to pull from, the human factor is surprisingly
absent with the internal list including “logs, networks, endpoints, malware, phish-
ing e-mails, and past incidents” and the external list including “industry sharing
groups, government, organization partnerships, vendors, and open source (Nelson,
2016).” Granted social media can be categorized under open source but the neces-
sity for it to become its own entity is becoming increasingly apparent. SANS sur-
vey results (Shackleford, 2015) show that tools were used by CTI professionals to
“aggregate, analyze, and present CTI” and that 76% of intelligence information
gathers from the security community. The results highlight that social media is still
an untapped resource although the demand exists.
Attempts to understand, dissuade, deter, and stop a hacker via social media
intelligence gathering could present an alternative solution to wading through the
wealth of attacks that lead to endless mazes of data. The study of hacker motivations
behind attacks may be limited because CTI falls directly into the realm of infor-
mation technology and computer science whereas the human criminal factor typi-
cally falls under the umbrella of law enforcement or military intelligence. Human
intelligence (HUMINT), for example, is a field that is not necessarily focused on
computers or advanced technology. Rather, HUMINT typically focuses on who is
416 ◾ Analytics and Knowledge Management

behind the latest national threat. Cybercrime is just now coming into mainstream
focus for national threats. HUMINT means deriving intelligence from people
rather than computers and an investigation usually begins with an event. It is the
event that drives the intelligence professional to discover the five Ws of who, what,
when, where, and why with the “why” usually being the least of concern due to
the fast-paced environment of threats. There is a relatively new field that focuses
more on the “why” and human behavior called sociocultural research. However,
the focus of this research, in general, is societies and the objective of the study is to
understand how and why society allows the threat to operate in that space. Social
media analysis could fall under the umbrella of all source intelligence (OSINT) but
it is still not designed specifically to look at those behind computer-related activity
and this type of analysis would be limited if not used in conjunction with CTI. In
summation, the study of social media to predict cyber threats by analyzing human
motivations is really a field or area of expertise that has yet to be established.

Benefits of Social Media Analysis in Cyber Threat Intelligence

Tracking hackers via regular every day activity, country environments, history, edu-
cation, and social contact has the potential to narrow the scope of investigation
compared to tracking from the end of the attack. To put this in terms of scale, there
are approximately 7.5 billion people in the world, 3.7 billion of whom are internet
users (Internet World Stats, 2017). The average ratio of criminal or crime type per
capita is under 10% (Nation Master, 2002). This means the plausible number of
potential criminals among internet users is less than 3.7 million users. To establish
capability, the ratio of user to technician is estimated to be anywhere from 50:1
to about 80:1 (National Center for Education Statistics, 2000; Rumberg 2012),
which means out of 3.7 million internet users with the possible intent to conduct
criminal activity, there is a good chance that only 46,000–74,000 have the actual
capability of being a hacker. These numbers can be further diminished by nar-
rowing the scope to countries of interest as one research study showed that most
hacks came from the United States and China (Kumar and Carley, 2016). The
global threat intelligence report from Dimension Data states that 63% of attacks
based on IP addresses originate from the United States and the top five countries
for attack origin account for 75% of all attacks in 2016 (Thomas, 2017). Many of
these attacks will be state-sponsored so the concern for a state would be either an
out of state hacker or a nonsanctioned in-state hacker which narrows the field of
potentials even further. Although these are broad estimations, these numbers are
a lot more manageable than tracing hackers from attack end points. For example,
a single threat tracker found almost 4 million attacks in one day, on July 2, 2017
(ThreatCloud, 2017). The benefit of tracking humans as opposed to attacks is the
number of possibilities is substantially smaller and will not change as drastically the
way cyberattacks morph by the day. Just these estimates create a good incentive to
track computer operators as a form of threat mitigation.
 Data Analytics for Cyber Threat Intelligence ◾ 417

Cyber threat intelligence (CTI) is a growing area in the intelligence community

which is occurring at unprecedented rates to secure technology vulnerabilities that
have taken the lead ahead of those responsible for securing it. Establishing a struc-
tured and defined trade in CTI has proven to be a colossal task as the very basics for
defining the field, like qualifications standards, common taxonomies for technical
language, and formalized trade techniques, are unstructured. As if transitioning the
intelligence tradecraft to CTI is not daunting enough, the use of Big Data analyt-
ics is still in its infancy in the intelligence world. Meanwhile, incorporating social
media into that CTI analysis is an even more untapped resource. The New York
Times recently reported that hackers are now embedding attacks into social media
recognizing it as an easier gateway into secured systems (Frenkel, 2017). As this
field develops, the role of algorithms that support text analysis will grow in demand
since the overwhelming amount of data will require simplified solutions to sort
through the maze.

Establishing Motivations in Cyber Threat

Intelligence through Social Media Analysis
Indications that establishing hacker motivations is growing in relevance is seen in
CTI analysis but is usually portrayed from the perspective of cyberattacks. The
image in Figure 13.2 shows an example of how hacker motivations (Hackmageddon,
2017) are typically displayed or understood. Humans, however, are complex indi-
viduals and motivations are often much more nuanced than what appears at face
value. For example, kidnappings can be a tool to generate income. However, kid-
nappings, although classified as crime, are often used as a functional tool to meet
a variety of objectives. Motivations driving this tool can vary greatly from love,
desire, revenge, control, religious beliefs, cultural practices, political influence, or
economics (ThreatRate Risk Management, 2017).
To clearly understand the motivations behind an attack, the mind of the
attacker should be analyzed and understood. This can be done by looking at
human behavior. Considering the operator exists in the cyber world, insight into

3.50% 1.20%

21.20%
Cyber crime
Cyber espionage
74.10%
Hactivitism
Cyber warfare

Figure 13.2 Motivations behind cyberattacks.

418 ◾ Analytics and Knowledge Management

the minds of the operator using Big Data analysis and text analysis is a logical
method of CTI collection. These operators, like all humans, have a basic human
desire for social interaction. Granted, there may be anomalies to this but it is safe
to assume that if an operator lives in the world of cybercrime, then the operator’s
social interaction would likely take place in the cyber world as well. Using text
analysis of social media could thus be an effective strategy for narrowing and inter-
cepting the threat.
Understanding the motivations behind the actions of an intruder also offers the
potential to provide alternative means for mitigating threat as each human moti-
vation is satisfied in different ways. There have been numerous studies regarding
the complexity of motivations (Marsden, 2015) and how varied human motiva-
tions can be. The Department of Defense is realizing the advantages of behavioral
analysis in detecting intrusion, which is why it is pushing for a new acquisition
model that would allow them to invest in private sector machine learning tech-
nologies (Owens, 2017). As strategic politics and war move into the digital era
and away from conventional techniques, the development of this type of intel-
ligence may become more and more relevant, especially as governments begin to
lose traditional forms of power and control with the proliferation of information
dissemination.

Role of Behavioral and Predicative Analysis in Cyber

Threat Intelligence
Scales are often used in behavioral analysis to diagnose human and animal behav-
ior. For example, Bech, Hamilton, and Zung developed a social dysfunction and
aggression scale, which allows a person’s level of aggression to be rated from irrita-
bility to physical acts of violence (Bech, 1996). This type of scale can be beneficial
in establishing a threat level in a work environment. Freedom House developed
the political terror scale seen in Table 13.2, which rates the level of political terror
across a country. This type of scale assists in identifying dictatorial or corrupt rule
and assists in decision making for international intervention or sanctions but more
importantly it can be used as a warning for potential government collapse or civil
war. Scales like these are useful in predictive analysis in a multitude of other fields
from economic trends to the development of social issues.
Scales in behavioral analysis can also be used to assign weight to words or
actions. Weights help the researcher identify the development or trajectory of
behavior. A classic example is the scale used by canine behaviorist to monitor the
transition of aggression in canines: (biting, snapping, growling, stiffening up,
lying down, standing crouched, creeping, walking away, turning body away, turn-
ing head away, and yawning). In this ladder, the development of canine aggressive
behavior is demonstrated in the form of a ladder. The bottom of the ladder begins
with subliminal signs of being uncomfortable that are obvious to dogs but are often
ignored by humans. The top of the ladder ends with overt signs of aggression that
 Data Analytics for Cyber Threat Intelligence ◾ 419

Table 13.2 The Political Terror Scale

Level Interpretation

1 Countries under a secure rule of law, people are not imprisoned for
their views. Political murders are extremely rare.

2 There is a limited amount of imprisonment for nonviolent political

activity. However, few people are affected and torture and beatings
are the exception. Political murder is rare.

3 There is extensive political imprisonment, or a recent history of such

imprisonment. Execution or other political murders and brutality
may be common. Unlimited detention, with or without a trial, for
political views is accepted.

4 Civil and political rights violations have expanded to a large number of

the population. Murders, disappearances, and torture are a common
part of life. In spite of its generality, on this level terror affects those
who interest themselves in politics or ideas.

5 Terror has expanded to the whole population. The leaders of these

societies place no limits on the means or thoroughness with which
they pursue personal or ideological goals.
Source: The Political Terror Scale, Documentation: Coding Rules, http://www.
politicalterrorscale.org/Data/Documentation.html#PTS-Levels, 2017.

are more familiar to humans but once displayed are often also ignored and usually
end in a bite. Canine behaviorists understand that if the problem can be identified
at its onset, mitigating action can take place to prevent the aggressive behavior.
Scales like the example in the canine ladder of aggression (Horwitz and Mills,
2009) can be used in a similar way for threat identification and prevention using
social media. Once the scale or trajectory of behavior is understood, the analyst
can data mine for these key words, assign weight to the words like scales and then
focus on the words that carry the most weight to prevent a hack. For example, if
the objective is to save lives by preventing dog bites, a canine analyst would moni-
tor behaviors in the middle of the aggression scale. Likewise, with human behavior,
if the goal is to diminish potential hacks, an analyst can search for words found
in the middle of the identified hacking “scale of aggression.” Both will narrow the
search for subjects that may commit the unwanted behaviors. The scope can be
narrowed even further by focusing on indicators that the behavior has yet to occur,
but the subject being reviewed has the potential or drive to commit the offense. In
the case of hackers, priority of intervention would then be placed on text indicating
the individual is in a current state of danger of performing a hack. Logically, this
method will be extended to CTI using the information available from social media.
420 ◾ Analytics and Knowledge Management

The same technique can be applied to hackers using the language of a hacker.
Hypothetically, if each phase of the Cyber Kill Chain requires a separate set of
tasks, the words assigned to these tasks could potentially indicate where in the
Cyber Kill Chain the hacker is postured. The weight of the words used could pro-
vide warning or indication of movement to the next stage and thus narrow the
scope of intervention. The Office of the Director of National Intelligence developed
a common cyber threat framework that defines the actions of hackers in each stage
of the Cyber Kill Chain (Sweigert, 2017). What is missing and could be developed
is a hacker dictionary with the words associated to the most logical stage of the
framework. There is already a preliminary work titled “The Hacker’s Dictionary”
that tracks the lexicon of hackers (Raymond, 2017). After the words are assigned to
the stages, they could then be given weight and priority based on where the words
fall in the Cyber Kill Chain.

Text Analysis Tools

The field of text analysis is a rapidly growing type of research. The reason for this
recent surge in growth is because an increase in computing power allows for the
analysis of extremely large sets of data. Linguistic analysis is now becoming a com-
puter science topic just as much as it is a social science. By using linguistic analysis,
it is possible to develop machine-learning algorithms that do tasks such as auto
completion or even smarter machines. Another use for language analysis is that
it can give insight into the thinking of the writer. It is for this reason that many
researchers are using social media outlets as a source of data. By analyzing the text
on social media, it is possible for researchers to get a feel of how large groups of
people act regarding certain topics like politics.
Textual analysis is the act of making an educated guess at the meaning of that
text. By analyzing text from various aspects of life such as television, films, and
magazines it is possible to get an understanding of the way certain cultures view
and interact with the world around them. Currently the largest repository of text
can be found on the internet, which means that it is useful to have tools that allow
us to use and manipulate digital text. The use of these text analysis tools is referred
to as computer assisted text analysis (Gee, 2015).
Analyzing written language is called text analysis, also referred to as text min-
ing, and involves analyzing text-based content using natural language processing
(NLP). Natural language processing is the ability of computers to understand what
users are writing in human writing patterns. This means that the computer must be
capable of identifying various dialects, slang, and even properly identify homonyms.
To allow computers to better understand human language, most NLP algo-
rithms are based on statistical machine learning, using statistical inferences. These
inferences come from rules gained from analyzing large sets of documents know as
corpora, that contain the correct values that need to be learned.
 Data Analytics for Cyber Threat Intelligence ◾ 421

Linguistic Inquiry Word Count

Linguistic analysis is the scientific analysis of a language sample and is a form of tex-
tual analysis. A language sample can be spoken or written language. By analyzing
the language sample, it is possible to gain an understanding of the writer. Linguistic
analysis has 5 branches which are phonology, morphology, syntax, semantics, and
pragmatics. Phonology is concerned with determining which speech sounds con-
trast with one another to produce differences in meaning and which speech sounds
are in complementary distribution and never produce meaningful contrasts.
Morphology is the identification and study of morphemes, a minimal part of a
word that cannot be further subdivided. Syntax is the structure that sentences must
follow to make sense in that given language. Semantics is the study of the meaning
of a word, including its connotation and its denotation. Pragmatics is how humans
use linguistic cues in actual situations.
Linguistic inquiry word count (LIWC) is text analysis software. It is used to
gain insights about the mood of the writer. The research for it began from a study
done by Pennebaker and Beall in 1986. The study discovered that when people were
asked to write about emotional moments in their life, there was evident improve-
ment in their physical health. To determine the correlation between these two
things, a panel of judges was assembled and ask to read emotional essays and rate
them among categories such as how coherent it was, how well it was organized,
emotional words used, and others. The issue that was found with this approach is
that most judges could not agree on the ratings of the papers in these categories.
Also, it was noticed that when the judges read emotional essays it had an effect on
their moods, which could impact on their ratings of the work.
It is for this reason that LIWC was created. LIWC has two primary compo-
nents: the processing component and the dictionary component. The processing
component uses natural language processing to read in the text file. The dictionary
component is a user-defined dictionary that is filled with words. Each word in the
dictionary is categorized. There are 64 categories overall and each word can fall into
more than one category. These categories can be either subjective, such as happy
and sad, or objective, such as nouns and verbs.
As LIWC goes through the text file, it compares the words from the text file to
the words in the dictionary file to determine their category. If a word in the text
file matches a word in the dictionary file, all the categories that the word belongs
to are incremented.
Even though LIWC currently has more than 80 categories, when it was first
developed it started with two broad categories, which are still important within the
program. These two categories are style words and content words. Content words
are words such as nouns and regular verbs. The purpose of these words is to convey
the meaning of the communication. Style words are words such as pronouns or
prepositions, which reflect how people communicate, as opposed to content words,
which reflect what people are saying.
422 ◾ Analytics and Knowledge Management

Since LIWC is a major tool used in the text analysis field there have been
experiments done to determine how effective the program is. At the University of
Illinois (Kahn et al., 2007) three experiments where done to test the effectiveness
of LIWC. In the first experiment, users where asked to write an amusing essay and
an emotional essay. LIWC was then used to analyze these essays and determine the
emotions the author was attempting to convey in them. LIWC was successful at
this and was able to identify the positive word usage in the amusing essay as well
as the negative emotions being conveyed in the emotional essay. During the second
experiment, users’ emotions were manipulated using film clips. After the users were
shown these clips, they were asked to give an oral presentation about how they felt.
This oral presentation was transcribed into text and analyzed using LIWC. The
results were accurate and they demonstrated that LIWC could correctly pick up
various traits from one text. The third experiment was similar to the second experi-
ment only this time the users were asked in-depth questions that helped determine
more about their actual personality. They were then once again shown clips similar
to those in the second experiment. The purpose of this experiment was to deter-
mine if LIWC could actually tell the difference in users’ personalities based on the
words they used. The results of this application, that LIWC could not tell the dif-
ference, reflected a more challenging scenario.
There have been several ways in which people have sought to use sentiment
analysis. One of the most popular ways has been in prediction of elections. Twitter
is a medium allowing for microblogs up to 140 words in length, known as “Tweets,”
to be posted by users. Other users on the site then read these Tweets and have the
option of retweeting them. Tumasjan et al. (2010) analyzed over 10,000 messages
that contained references to a political party of politician during the German fed-
eral election. The purpose of this analysis was to determine whether you could use
Twitter as a valid medium for determining the offline political landscape in regards
to the public’s opinion.
This study was based off the fact that many believe that the current United
States president’s victory was due to his ability to effectively use social networks
in his campaigning. The president not only used social networking but he even
created his own website to increase his online presence. This led the researchers to
wonder to what extent social networks play roles in politics and whether it was an
effective place for gathering information on what the public currently felt about the
political parties and the politicians in those parties. The researchers also wanted to
determine whether Twitter was a platform for political deliberation, whether it was
an accurate reflection of political sentiment, and whether it could accurately predict
election results.
After examining 104,003 political Tweets regarding the 6 political parties that
were part of the election or Tweets that mention prominent politicians in those par-
ties, the researchers were able to answer these questions. Regarding whether Twitter
is a platform for political deliberation, it was found that there was a high amount
of political debate on Twitter; however, it was coming from a small group of heavy
 Data Analytics for Cyber Threat Intelligence ◾ 423

users, which may lead to an unequal representation overall. Regarding the second
question, it was found that the examination of users’ Tweets about the political par-
ties and politicians did show that users could accurately identify certain traits about
the moderate politicians and parties unless they were completely veered towered
one side of the political spectrum. Regarding the last question, there was a correla-
tion found between the number of Tweets where the party was mentioned and the
winner of the election.

Sentiment Analysis
Sentiment analysis is the use of NPL and text analysis to determine the attitude of
the speaker on the topic. The primary purpose of sentiment analysis is to determine
if a sentence or even a document is positive or negative; this is called polarity senti-
ment analysis. A common use for polarity sentiment analysis is on product reviews.
By analyzing the text of the review, it can be determined whether the review is
positive or negative.
There are numerous ways that have been proposed to conduct sentiment analy-
sis. Pang and Lee (2004) proposed a machine learning method that categorized the
subjective portions of the document. This method differs from previous methods
of classifying documents which was based on selecting lexical features of the docu-
ment such as the word good being present. Instead, the author proposed to only
use subjective sentences and not the objective sentences, which can be considered
to be misleading.
Web 2.0 has the chance for businesses to use sentiment analysis in more busi-
ness efficient matters. Web 2.0 is characterized by its increase in interactions
between users and web pages. Facebook posts, Twitter posts, and even product
reviews allow for analysis of large datasets, which can be used to create predica-
tion models. These predication models can be used to help companies determine
how successful their next product would be or what features they should include
in their future products. Another use for this analysis is to attempt outcomes of
governmental elections.
When attempting to determine the sentiment of writing one of the issues that
may cause skewed results is the misreading of the text into the analysis software.
One of the foundations of sentiment analysis is polarity. The polarity is either nega-
tive or positive. An example of this can be seen in application reviews, which can
either have a positive review (thumbs up) or a negative review (thumbs down).
When attempting to determine the polarity of a document there have been
many approaches purposed. One of the most common methods is to select lexical
features like indicative words such as “good.” Upon seeing “the word good” in the
text of a review of a novel, many analysis tools may identify that as a positive polar-
ity which may skew the overall results even though the sentence where it was used
had nothing to do with the reviewers feeling about the actual novel. For example,
the reviewer may write “The lead character attempted to protect his good name.”
424 ◾ Analytics and Knowledge Management

It is for this purpose that Pang has proposed the use of a minimum cuts method,
which separates the subjective and objective sentences from text and only considers
subjective sentences in the analysis phase. They propose a document level subjectiv-
ity detector, which will allow for subjectivity detection on the sentence level. Only
sentences that passed through the subjectivity detector will be allowed in to the
polarity classifier for document analysis.
The results of applying this subjectivity detector are that they gained a statisti-
cally significant improvement of 4% in the accuracy of the sentiment analysis of the
given documents.
The reason that Twitter is such a popular place for sentiment analysis is due
to the high value of text that is present there as well, how easy it is to gather the
text, the wide variety of users it has, and the possibility to collect data from other
cultures. This offers a plethora of data for companies such as what is the best demo-
graphic for their product and how the public reacts to certain forms of marketing.
To demonstrate how effective Twitter can be for a corpus of text for sentiment
analysis, Pak and Paroubek (2010) collected 300,000 texts from Twitter, which
they then separated into three groups evenly. The categories were: text that con-
veyed positive emotions, text that conveyed negative emotions, and objective text.
Linguistic analyses were then collected on this corpus and a sentiment classifier was
then developed using the corpus as training data.
The corpus was collected using the Twitter API that allows you to actually
query a collection of Tweets from the website. In the query, happy emoticons
and sad emoticons were searched. Tweets containing sad emoticons where con-
sidered negative text and tweets using positive emoticons where considered posi-
tive text. Objective text was pulled from Tweets of magazines and newspaper such
as The New York Times. The classifier was built using a multinomial naive Bayes
classifier with binary n-grams from the text as the input.

SentiStrength
SentiStrength is a tool for sentiment analysis. Textual sentiment analysis tools
focus largely on determining positive, negative, or lack of, sentiments within
the given text. To accomplish this, tools either use a machine-learning or lexical
approach to analyzing the text. The machine-learning method works by using
n-grams. The software is trained to mark certain n-grams as positive and others
as negative. The lexical approach is similar to that of LIWC where the software
is given a dictionary file that identifies certain words as negative or positive, and
then assigns a value during the analysis of the text. The lexical approach also
uses pseudo n-gram styles such as the word “not”; it is read as a negation of the
following word, so the words “not happy” would be read as a negative statement
(Thelwall, 2013).
SentiStrength is a sentiment analysis tool that uses the lexical approach for
analysis. There are three versions of SentiStrength: the online version, the free Java
 Data Analytics for Cyber Threat Intelligence ◾ 425

version which is made available for research and educational purposes, and the
paid version which is also written in Java, but made for commercial use. All the
versions contain similar functions but the commercial version offers the ability to
scan and organize much larger datasets.
Unlike LIWC, which is popular due to its dictionary, the main draw to
SentiStrength is the algorithm it uses to determine the sentiment in a document.
As mentioned earlier, SentiStrength is a lexical sentiment analysis tool; this means
that it uses a dictionary to determine the sentiment of words in the given text. The
dictionary file is made up of a combination of the LIWC dictionary but has been
updated periodically through testing. In the dictionary file for SentiStrength, each
word is given a positive sentiment score of 1 to 5 and a negative sentiment score of
−1 to −5. A score of 1 means no positive found and a score of −1 means no negative
emotion was found. The sentiment of the overall sentence is then given the senti-
ment score of the highest positive and negative sentiment in that statement
Textual analysis is the analysis of text to gain useful information from it.
Computer-assisted textual analysis done by computer programs using natural lan-
guage processing which allows computers to convert high-level human language
into a computer-readable format. Many NLP programs also use statistical analysis
within their algorithm to be predictive of text that is being read in.
The difference between LIWC and sentiment analysis is that linguistic analy-
sis is a form of textual analysis that is done by using certain language structures
found in most languages to get an understanding of the text. There are a variety
of linguistic analysis tools available, but one of the most well used tools is LIWC.
The two main parts of LIWC are the processing component and the dictionary.
The processing component is what reads in the text file and it is then compared
to the dictionary component and sorted based on the category of the word in the
dictionary. By doing this, LIWC can then find patterns within those categories that
may give information about the meaning behind the writing, such as whether it is
joyous or sad. Sentiment analysis is similar to linguistic analysis in what it seeks
to accomplish. The notable difference between the two is that linguistic analysis is
attempting to understand the language that is being used in the writing whereas
sentiment analysis is attempting to determine the overall polarity of the document.
With the rise of social media, sentiment analysis is being used to analyze large sets
of text to determine certain trends found in it.

Case Study Using Linguistic Inquiry Word Count

To better combat cyber threats, many researchers (Chi et al., 2016; SANS, 2017)
have begun to focus more on the psychology of insiders who become threats and
what causes them to do so. Collecting data from targeted personnel, such as social
networking and emails, is the best way to prevent or identify an insider threat early.
Text analysis will analyze that data and give policy makers the resilient decision
support.
426 ◾ Analytics and Knowledge Management

Text data is analyzed using LIWC to determine if a given actor is a threat. The
difference between the past methods of determining an insider threat is that there
has never been a study that used linguistic analysis to bridge the gap between an
employee’s text and the demonstration that they could possess the characteristics
of the dark triad model. The data is tested for agreeableness, neuroticism, conscien-
tiousness, and extraversion. Since both agreeableness and conscientiousness are traits
consistent with all three of the dark triad traits, this is the first pair that is looked
for in the actor. The LIWC scores in the required categories are tested against the
average of known insider threat cases to determine if the actor possesses these traits.
If these traits are identified, the actor’s text is tested against the LIWC catego-
ries for neuroticism and extroversion. Depending on the scores for these categories
it becomes evident if the actor does or does not possess the characteristics of these
mental diseases. If the actor does possess the traits for one of the dark triad person-
ality traits, then they are considered to be a possible insider threat.
Of the three dark triad traits, this is the first pair that is looked for in the actor.
The LIWC scores in the required categories are tested against the average of known
insider threat cases to determine if the actor possesses these traits. Figure 13.3 rep-
resents the method by which threat values will be assigned to users.
The underlying problem, when determining whether or not the psychology of
a person makes them more likely to pose a threat to the company, is if their actions
make them less trustworthy. Trust is a concept, which is multidimensional and effects
relationships on both small and large scales. When an organization hires an individ-
ual, it is important that they trust that individual. It is for this reason that many com-
panies give new employees personality evaluations. These personality questions are

• Text is analyzed using LIWC

• The resulting numbers are
Linguistic used to determine the FFM
analysis Traits as well as the dark
triad traits

Identify • The MAD algorithm is used to

outliers identify outliers

• Determine the values of the scores

Assigning a from the FFM and dark triad
threat level model for outliers
• If more than 3 of the values are
greater than the risk, value is HIGH

Figure 13.3 Threat assessment method.

 Data Analytics for Cyber Threat Intelligence ◾ 427

based on the traits identified in the five factor model (FFM) (Judge and Bono 2000).
FFM is a widely accepted model that is used to identify certain personality traits in
people based on certain mental cues. The problem with this personality test is that
many people lie to get hired by companies. Response distortion is common among
many applicants; therefore, many companies may not get an accurate understanding
of an employee’s true personality.
The outliers for social networking after LIWC analysis can be identified using
median absolute deviation (MAD). MAD is a method to identify outliers. The
MAD (Leys et al., 2013) was calculated as shown below. Here M is the median and
x the values:

MAD = 1.4826 * M i (| xi − M j ( x j )|)

The decision criteria for the outliers are as shown below. We chose mi = 2.5. A score
of 3 is very conservative, 2.5 is moderately conservative, and 2 is poorly conservative.

Conclusions
This chapter gives an overview of data analysis uses in the field of CTI and insider
threat. All data collection resources are related to mobile devices. This chapter exam-
ined the process of collecting and organizing data, various tools for text analysis, and
several different analytic scenarios and techniques. Applying data analytics into the
field of CTI is still in its infancy. Meanwhile, incorporating social media into CTI
analysis is an even more untapped resource (Clark, 2016). Dealing with a collec-
tion of very huge datasets with a great diversity of types from social networks, data
analytic provides enormous novel approaches for CTI professionals. How to use Big
Data analytic technique to achieve valuable information is still big challenge.

References
Barnum, S. (2014). Standardizing Cyber Threat Intelligence Information with the Structured
Threat Information eXpression (STIX™), Mitre. February 20. Version 1.1, Revision 1.
http://stixproject.github.io/about/STIX_Whitepaper_v1.1.pdf. Accessed June 26, 2017.
BBC Europe (2017). Cyber-attack: Europol says it was unprecedented in scale. BBC News,
May 13. http://www.bbc.com/news/world-europe-39907965. Accessed June 23, 2017.
Bech, P. (1996). The Bech P., Hamilton and Zung Scales for Mood Disorders: Screening and
Listening. Springer-Verlag: Berlin, Germany.
Bennett, C. (2015). Russian hackers crack pentagon email system. The Hill, August 6.
http://thehill.com/policy/cybersecurity/250461-russian-hackers-crack-pentagon-
email-system. Accessed July 8, 2017.
Bronk, C. (2016). Cyber Threat: The Rise of Information Geopolitics in U.S. National Security.
Praeger: Santa Barbara, CA.
428 ◾ Analytics and Knowledge Management

Castells, M. (2007). Communication, power and counter-power in the network society.

International Journal of Communication 1(1): 29.
Chi, H., Scarlett, C., Prodanoff, Z.G., and Hubbard, D. (2016, December). Determining
predisposition to insider threat activities by using text analysis. Future Technologies
Conference (FTC) (pp. 985–990). IEEE.
CIA (2000). Statement for the Record Before the Joint Economic Committee on Cyber threats and
the US Economy by John A. Serabian, Jr. Information Operations Issue Manager, CIA.
Central Intelligence Agency News and Information: Cyber Threats and the U.S. Economy.
https://www.cia.gov/news-information/speeches-testimony/2000/cyberthreats_022300.
html. Accessed June 21, 2017.
Clark, R.M. (2016). Intelligence Analysis: A Target-centric Approach. CQ Press: Washington, DC.
CNN Wire Staff (2011). Obama says U.S. has asked Iran to return drone aircraft. CNN,
December 13. http://edition.cnn.com/2011/12/12/world/meast/iran-us-drone/index.
html. Accessed June 23, 2017.
Davis, J. (2015). Hacking of government computers exposed 21.5 million people. New York
Times, July 9. https://www.nytimes.com/2015/07/10/us/office-of-personnel-
management-hackers-got-data-of-millions.html. Accessed June 23, 2017.
Davis, K. (2016). Phantom and ESG research finds companies ignore majority of secu-
rity alerts. Business Wire Inc., March 15. http://www.businesswire.com/news/
home/20160315005555/en/Phantom-ESG-Research-Finds-Companies-Ignore-
Majority. Accessed July 9, 2017.
The Department of Defense Cyber Strategy (2015). https://www.defense.gov/Portals/1/
features/2015/0415_cyber-strategy/Final_2015_DoD_CYBER_STRATEGY_for_
web.pdf. Accessed July 24, 2017.
Department of Justice (2007). Analyst toolbox: A toolbox for the intelligence analyst.
Global Justice Information Sharing Initiative. https://it.ojp.gov/documents/analyst_
toolbox.pdf. Accessed July 20, 2017.
Dickson, B. (2016). How predictive analytics discovers a data breach before it happens.
Crunch Network, July 25. https://techcrunch.com/2016/07/25/how-predictive-
analytics-discovers-a-data-breach-before-it-happens/. Accessed July 9, 2017.
DOD Defense Science Board (2013). Task Force Report: Resilient Military Systems and the
Advanced Cyber Threat. Office of the Under Secretary of Defense for Acquisition,
Technology and Logistics: Washington, DC, January. http://nsarchive.gwu.edu/
NSAEBB/NSAEBB424/docs/Cyber-081.pdf.
Edwards, C., Migues, S., Nebel, and Owen, D. (2002). System and method of data collec-
tion, processing, analysis, and annotation for monitoring cyber-threats and the noti-
fication thereof to subscribers. https://www.google.com/patents/US20020038430.
Accessed June 23, 2017.
Farham, G. (2017). Tools and standards for cyber threat intelligence projects. SANS Institute
InfoSec Reading Room, October 14. https://www.sans.org/reading-room/whitepapers/
warfare/tools-standards-cyber-threat-intelligence-projects-34375.
Frenkel, S. (2017). Hackers hide cyberattacks in social media posts. New York Times, May 28.
https://www.nytimes.com/2017/05/28/technology/hackers-hide- cyberattacks-
in-social-media-posts.html. Accessed July 2, 2017.
Gee, J. (2015). Social Linguistics and Literacies: Ideology in Discourses. Routledge: New York.
Gragido, W. (2012). Understanding Indicators of Compromise (IOC) Part I, October 3, 2012.
https://blogs.rsa.com/understanding-indicators-of-compromise-ioc-part-i/. Accessed
July 7, 2017
 Data Analytics for Cyber Threat Intelligence ◾ 429

Hackmageddon (2017). Motivations behind attacks. Information Security Timeline and Statistics.
http://www.hackmageddon.com/2017/06/09/april-2017-cyber-attacks-statistics/.
Hayes, N. (2016). Why social media sites are the new cyber weapons of choice. Dark Reading,
September 6. http://www.darkreading.com/attacks-breaches/why-social-media-sites-
are-the-new-cyber-weapons-of-choice/a/d-id/1326802. Accessed July 9, 2017.
Horwitz, D., and Mills, D. (2009). BSAVA Manual of Canine and Feline Behavioral
Medicine. BSAVA: Gloucester, UK.
Hutchins, E., Cloppert, M., and Amin, R. (2011). Intelligence-Driven Computer Network
Defense Informed by Analysis of Adversary Campaigns and Intrusion Kill Chains.
Lockheed Martin Corporation. http://www.lockheedmartin.com/content/dam/
lockheed/data/corporate/documents/LM-White-Paper-Intel-Driven-Defense.pdf.
Accessed July 8, 2017.
Internet World Stats (2017). http://www.internetworldstats.com/stats.htm. Accessed July 24, 2017.
Judge, T.A., and Bono, J.E. (2000). Five-factor model of personality and transformational-
leadership. Journal of Applied Psychology 85(5): 751.
Kahn, J.H., Tobin, R.M., Massey, A.E., and Anderson, J.A. (2007). Measuring emotional
expression with the linguistic inquiry and word count. The American Journal of
Psychology 120: 263–286.
Koerner, B. (2016). Inside the cyberattack that shocked the US government. Wired,
October 23. Condé Nast: New York, NY. https://www.wired.com/2016/10/inside-
cyberattack-shocked-us-government/. Accessed June 22, 2017.
Kumar, S., and Carley, K. (2016). Approaches to Understanding the Motivations Behind Cyber
Attacks. Carnegie Mellon University. http://www.casos.cs.cmu.edu/publications/
papers/2016ApproachestoUnderstanding.pdf. Accessed July 6, 2016.
Lapowsky, I., and Newman, L.H. (2017). Wikileaks CIA dump gives Russian hacking
deniers the perfect ammo. https://www.wired.com/2017/03/wikileaks-cia-dump-
gives-russian-hacking-deniers-perfect-ammo/. Accessed July 24, 2017.
Larson, S. (2017). NSA’s powerful Windows hacking tools leaked online. CNN Technology,
April 15. http://money.cnn.com/2017/04/14/technology/windows-exploits-shadow-
brokers/index.html. Accessed June 23, 2017.
Lee, R. (2014). Cyber threat intelligence. Tripwire, October 2. https://www.tripwire.com/
stateof-security/security-data-protection/cyber-threat-intelligence/. Accessed July 7, 2017.
Leys, C., Ley, C., Klein, O., Bernard, P., and Licata, L. (2013). Detecting outliers: Do not
use standard deviation around the mean, use absolute deviation around the median.
Journal of Experimental Social Psychology 49(4): 764–766.
MalwareInt (2017). Malware Tech Botnet tracker. https://intel.malwaretech.com/. Accessed
July 11, 2017.
Marsden, P. (2015). The science of why. Brand Genetics, June 22. http://brandgenetics.com/
the-science-of-why-speed-summary/. Accessed July 12, 2017.
Nakashima, E., and Krebs, B. (2007). Contractor blamed in DHS data breaches. The
Washington Post, September 24. http://www.washingtonpost.com/wp-dyn/content/
article/2007/09/23/AR2007092301471.html?hpid=sec-tech. Accessed June 22, 2017.
National Center for Education Statistics (2000). Technology in schools. Chapter 5:
Maintenance and Support, Technology in Schools: Suggestions, Tools, and Guidelines
for Assessing Technology in Elementary and Secondary Education. https://nces.ed.gov/
pubs2003/tech_schools/chapter5.asp. Accessed July 11, 2017.
NationMaster (2002). http://www.nationmaster.com/countryinfo/compare/Ireland/United-
States/Crime. Accessed July 25, 2017.
430 ◾ Analytics and Knowledge Management

Nelson, M. (2016). Threat Intelligence Capability. North Dakota State Capability. https://
www.ndsu.edu/fileadmin/conferences/cybersecurity/Slides/Nelson-Matt-Threat_
Intel_Capability_Kick_Start_.pptx. Accessed July 14, 2017.
OASIS (2017). OASIS cyber threat intelligence (CTI) TC. https://www.oasis-open.org/
committees/tc_home.php?wg_abbrev=cti. Accessed June 26, 2017.
Owens, K. (2017). Army uses behavioral analytics to detect cyberspace invaders. https://
defensesystems.com/articles/2017/06/29/army-cyber.aspx. Accessed June 29, 2017.
Pak, A., and Paroubek, P. (2010, May). Twitter as a corpus for sentiment analysis and opin-
ion mining. In LREc (Vol. 10). Université de Paris-Sud: France.
Pang, B., and Lee, L. (2004, July). A sentimental education: Sentiment analysis using
subjectivity summarization based on minimum cuts. In Proceedings of the 42nd
annual meeting on Association for Computational Linguistics (p. 271). Association for
Computational Linguistics: Barcelona, Spain.
The Political Terror Scale (2017). Documentation: Coding rules. http://www.political
terrorscale.org/Data/Documentation.html#PTS-Levels. Accessed July 12, 2017.
Rawnsley, A. (2011). Iran’s alleged drone hack: Tough, but possible. Wired. https://www.
wired.com/2011/12/iran-drone-hack-gps/. Accessed June 23, 2017.
Raymond, E. (2017). The New Hacker’s Dictionary, 3rd ed. The MIT Press. https://mitpress.
mit.edu/books/new-hackers-dictionary. Accessed July 11, 2017.
Rumberg, J. (2012). Metric of the month: Tickets per technician per month. MetricNet. http://
www.thinkhdi.com/~/media/HDICorp/Files/Library-Archive/Insider%20Articles/
tickets-per-technician.pdf. Accessed July 11, 2017.
SANS (2017). FOR578: Cyber threat intelligence. https://www.sans.org/course/cyber-
threat-intelligence. Accessed July 14, 2017.
SecureWorks (2017). Cyber threat basics, types of threats, intelligence & best practices.
https://www.secureworks.com/blog/cyber-threat-basics. Accessed July 9, 2017.
Sevastopulo, D. (2007). Chinese hacked into Pentagon. Financial Times, September 3.
https://www.ft.com/content/9dba9ba2–5a3b-11dc-9bcd-0000779fd2ac?mhq5j=e2.
Accessed June 23, 2017.
Shackleford, D. (2015). Who’s Using Cyberthreat Intelligence and How? Sans Institute
InfoSec Reading Room. https://www.sans.org/reading-room/whitepapers/analyst/
cyberthreat-intelligence-how-35767. Accessed June 21, 2017.
Sweigert, D. (2017). A Common Cyber Threat Framework: A Foundation for Communication. Cyber
Threat Intelligence Integration Center—ONDI. https://www.slideshare.net/dgsweigert/
cyber-threat-intelligence-integration-center-ondi?qid=4d6055a5-ead1–40d5–84b8–
60439c570852&v=&b=&from_search=6. Accessed July 14, 2017.
Tableau (2017). Top 10 big data trends for 2017. https://www.tableau.com/sites/default/
files/media/Whitepapers/whitepaper_top_10_big_data_trends_2017.pdf?ref=lp&sig
nin=66d590c2106b8d532405eb0294a4a9f1. Accessed July 6, 2017.
Thelwall, M. (2013). Heart and soul: Sentiment strength detection in the social web with
sentistrength. Proceedings of the CyberEmotions 5: 1–14.
Thomas, M. (2017). 2017 global threat intelligence report. Dimension Data, May 4. http://
blog.dimensiondata.com/2017/05/2017-global-threat-intelligence-report/. Accessed
July 9, 2017.
Thornburgh, N. (2005). Inside the Chinese hack attack. TIME Magazine, August 25. http://
content.time.com/time/nation/article/0,8599,1098371,00.html. Accessed June 23,
2017.
 Data Analytics for Cyber Threat Intelligence ◾ 431

ThreatCloud (2017). Live Cyber Attack Threat Map. Check Point Software Technologies Inc.
https://threatmap.checkpoint.com/ThreatPortal/livemap.html. Accessed June 3, 2017.
ThreatRate Risk Management (2017). Types of kidnappings. http://www.threatrate.com/
pages/47-types-of-kidnappings. Accessed July 6, 2017.
Thycotic (2017). PBA access use case. https://vimeo.com/209209431. Accessed July 11, 2017.
Tumasjan, A., Sprenger, T.O., Sandner, P.G., and Welpe, I.M. (2010). Predicting elections with
Twitter: What 140 characters reveal about political sentiment. ICWSM 10(1): 178–185.
U.S. Cong. Senate—Homeland Security and Governmental Affairs 113 Cong. (2014). Federal
Information Security Management Act of 2014. 113 Cong. S.2521. Washington, DC.
https://www.congress.gov/bill/113th-congress/senate-bill/2521. Accessed June 22, 2017.
U.S. Cong. House-Government Reform; Science. 107 Cong. (2002) Federal Information
Security Management Act of 2002. 107 Cong. 2nd Sess. H. R. 3844. Washington, DC.
https://www.congress.gov/bill/107th-congress/house-bill/3844. Accessed June 22, 2017.
Windrem, R. (2015). China read emails of top U.S. officials. NBC News, August 10. http://
www.nbcnews.com/news/us-news/china-read-emails-top-us-officials-n406046.
Index

Note: Page numbers followed by f and t refer to figures and tables respectively

A knowledge discovery in databases process,

32–33, 33f
Abookire, S., 128 lessons learned and future work, 398–400
Ackoff, R., 326 managerial, 7, 14–17
Acquisition, data, 208 methodology in student attrition study,
AdWords, 162 49–57, 51f, 51t–53t, 55f
AeroText, 213 methods best suited to goals, 371–375, 371t
Agent-based modeling, 288, 289f, 289t mixed methods analysis and variant sources,
Agora Plus, 214 374–375, 377t–381t
Algesheimer, A., 87 multiple project examples, 20–22, 21f
Allee, V., 225, 231 not supported by tools, 376–377
Alphabet, 348t operational, 8, 17–18
Alt, R., 204 operations research/management science,
Amazon, 18, 203, 205, 213 and business intelligence, 9–10
Amelia software, 19, 22 organizational environment and, 23–24
American Productivity and Quality Center political environment and, 24–25
(APQC), 3 predicting student attrition with, 46–57
Ammachchi, N., 137 predictive analytics models, 54–56, 55f
Amplified Analytics, 213 qualitative analysis and language based
Analytics, 2–5, 5f, 400–401 analytics, 372–374
Big Data, simulation modeling, and, quantitative analysis and, 371–372, 371t
287–293 research and business goals, 359–362,
cross-industry standard process for data 360t–370t
mining (CRISP-DM), 34–39, 34f, 43f sample, explore, modify, model, assess
customer-facing, 8, 18–20 (SEMMA) process, 39–42, 40f, 43f
development of, 8–9 scientific, 8, 20
discussion and conclusions, 59–63, 60t–61t, sensitivity analysis, 56–57
61f shift in practices and access, 144–146
elderly discharge planning simulation case, single-project examples, 12–20
291–292, 292t Six Sigma for, 42–45, 44f
evolution of, 201–203 social media. See Social media analytics (SMA)
examples overview, 10–12, 11t sources analyzed using quantitative data,
focal points of, 357, 358t 381–388, 383f, 388t–396t
future developments, 23–25 strategic, 7, 12–14
introduction to, 32–46 student attrition results, 57–59, 57t–59t

433
434 ◾ Index

Analytics (Continued ) Beekeepers, knowledge transfer practices

techniques, 208–211 among, 367t, 379t, 391t–392t
tools, 152–156, 176–177, 370–377, Behavioral analysis in cyber threat intelligence,
377t–381t 418–420, 419t
tools fitted to methods and sources, Bengston, D. N., 204
396–397, 398t Berger, T., 13
user feedback. See User feedback Berman, R., 384
which process is best in, 45–46, 46f Beyer, M. A., 284
workflow embedded in business processes, 25 BI&A. See Business intelligence and analytics
workflow models for text analytics, 386 (BI&A)
Analytics projects Biber, D., 384
categorizing, 6–8 Big Data, 2–3, 25, 142. See also Metrics
classification by technical type, 6–7 additions to simulation modeling, 290, 291f
classification from business perspective, 7–8 analytics tools, 152–156
multiple project examples, 20–22, 21f applications and case studies, 157–159
single-project examples, 12–20 characteristics of, 284–286, 286f
Analyze step, DMAIC process, 45 cost and anticipated value, 147–148
Anderson, M., 215 data type in, 168–169
ANN. See Artificial neural networks (ANN) data visualization and, 274
Apache Unstructured Information Management definitions of, 284, 285f
Applications (UIMA), 212 emerging career opportunities, 160–162
Appel, O., 125 information visualization, 148–151, 151f
Apple, 347, 348t intangibles and, 328–329
Arab Spring revolutions, 363t–364t, 377t, intelligence hierarchy in organizations and,
388t–389t 333–335
Area graphs, 258–260, 260f wider concept of knowledge management
Armano, D., 200 and, 143–144
ARPANET, 199 Big Data analytics (BDA), 2–3, 25
Arthur D. Little Co., 9 Biosciences, social media analytics in, 215
Artificial intelligence-assisted simulations, 283 BlackPlanet.com, 200
Artificial neural networks (ANN), 41–42, 54 Blei, D. M., 209
discussion and conclusions, 59–63, 60t–61t, Blogger, 200
61f Blogging sites, 200
results, 57–59, 57t–59t Blue Cross, 11t, 18
AsianAve.com, 200 Blue Hill Research, 7
Asilomar AI Principles, 25 Bokeh, 252
Assess step, SEMMA process, 42 Bollen, J., 214
Automated driving, intangible assets and Bootcamps, 191
metrics in, 347–351, 348t Booz Allen Hamilton, 9
Borgatti, S., 212
B Botha, A., 132
Bowerman, B. L., 92
Baidu, 348t Boxplots, 255–256, 256f
Banking to 45º, 272 Boyd, D., 199
Bank of America, 11t, 20 Brailsford, S. C., 288
Bar charts, 253–254, 254f Breiman, L., 56–57
Bar-Hillel, Y., 384 British Orthopedic Association, 297
Barker, A., 284 Brooker, P., 210
Barnett, J., 210 Brunelli, M., 10
Barton, D., 384 Bruns, A., 204
Batrinca, B., 198, 214 Bubble charts, 258, 259f
BBS. See Bulletin board system (BBS) Buffer, 214
 Index ◾ 435

Bulletin board system (BBS), 199 Computer-supported cooperative work

Burns and stalker, 77 (CSSW), 70
Business applications, 212 Computing, 2–3, 10
Business intelligence and analytics (BI&A), Condition-based maintenance (CBM), 13
9–10, 73–74, 201 Control step, DMAIC process, 45
Business perspective analytics projects, 7–8 CoP. See Communities of practice (CoP)
Business understanding, CRISP-DM, 35 Cope, B., 384
CoSchedule, 214
C Costco, 158
Coursera.com, 162
C++ (programming language), 180 Cribbin, T., 210
Cancer Survivors Network (CSN), 129 CRISP-DM. See Cross-industry standard
Capey, R., 8–9 process for data mining
Cap Gemini Consulting, 6, 23 (CRISP-DM)
Capital One, 162 Critical incident technique, 239–240
Carlsson, C., 10 Cross-industry standard process for data mining
Caterpillar Marine, 11t, 16, 22 (CRISP-DM), 34–39, 34f, 43f
Cellier, F. E., 283, 290 methodology in student attrition study,
Chambers, M., 3, 7, 10–12 49–57, 51f, 51t–53t, 55f
Chen, C., 13 as preferred process, 45–46, 46f
Chen, H., 9–10 CSCL. See Computer-supported cooperative
Chenoweth, M., 13 learning (CSCL)
Chiang, R. H. L., 9–10 CSSW. See Computer-supported cooperative
Christie hospital, 11t work (CSSW)
Clarabridge, 213 CTI. See Cyber threat intelligence (CTI)
Cleanup, data, 176 Customer engagement, 205–206
Cleveland, W. S., 170, 267, 269t–270t, 271–272 Customer-facing analytics, 8, 18–20
Clinton, Hillary, 158 Cyber threat intelligence (CTI)
Cloudera, 154, 162 benefits of social media analysis in, 416–417
Cluster-based flows of patients, modeling of, challenges in, 413–415
302–305, 303f, 303t–305t computational methods in, 411–413, 412t
Cluster-based system dynamics model, establishing motivations in, 417–418, 417f
303–305, 304t–305t introduction to, 408–409, 409f
Clustering approach, 298 linguistic inquiry word count (LIWC) in,
Clustering experiments, 298–300, 300f 421–427, 426f
Clusters, exploring, 300–301, 301f–302f prevalence of social media analysis in,
Coca-Cola, 9–10, 11t, 17 415–416
Cohen, G., 14 role of behavioral and predicative analysis
Collective knowledge systems (CKSs), 71 in, 418–420, 419t
Collective knowledge within communities of sentiment analysis in, 423–424
practice, 69–71 SentiStrength, 424–425
Color in data visualization, 271–272 significance of, 409–410
Colruyt, 11t, 19 text analysis tools, 420–427, 426f
Communities of practice (CoP) Cyworld, 200
collective knowledge within, 69–71 Czarniawska-Jerges, B., 77
Twitter-based communities as, 76–77,
115–116 D
CompuServe, 199
Computational social science, social media Dang-Xuan, L., 206
analytics in, 215 Data
Computer-supported cooperative learning acquisition of, 208
(CSCL), 70 cleanup of, 176
436 ◾ Index

Data (Continued) “Data Science: An Action Plan for Expanding

collection of, 175–176 the Technical Areas of the Field of
CRISP-DM preparation of, 36–37 Statistics,” 170
CRISP-DM understanding of, 35–36 Data scientists, 173–174
historical, 207 Data stream management systems (DSMSs),
identifying sources of, 207 211
integration of, 176 Data visualization. See Visualization, data
of public interest, 169 Data warehouse (DW) repositories, 74–75
real-time, 207 Davenport, T. H., 6, 9–10, 16, 168
scientific, 168 DBMSs. See Database management systems
social media, 169, 204–207 (DBMSs)
transaction, 169 Decision support systems (DSSs), 74
visualization of. See Visualization, data Decision trees, 54–55
web, 169 Deep neural networks (DNN), 318, 318f
Database management systems (DBMSs), 211 Deerwester, S., 210
Data-driven feedback loop, 281f Define, measure, analyze, improve, and control
Data-information-knowledge-wisdom (DIKW) (DMAIC) process, 42–45, 44f
hierarchy, 326 Define step, DMAIC process, 44
Data science. See also Big Data Delen, D., 38
analysis of jobs in data analytics and, Delphi, 348t
174–180, 181f De Mauro, A., 284
analytics tools, 152–156 Deployment, CRISP-DM, 39
applications and case studies, 157–159 DES. See Discrete event simulation (DES)
characteristics of employers in, 177–178, Descriptive analytics, 6
177f Dhar, V., 174
data analysis tools, 176–177 Dibble, E., 271
data cleanup and integration in, 176 Dimensional model, Twitter-based community
data collection in, 175–176 knowledge represented in, 78–81,
data, data types, and Big Data in, 168–169 79f–81f
defining, 170–171 Dinsmore, T. W., 3, 7, 10–12
education, 182–192 Discrete event simulation (DES), 288, 289f,
emerging career opportunities, 160–162 289t
graduate certificate programs in, 186 modeling patient’s care journey, 305–308,
information visualization in, 148–151, 151f 306f–308f
job titles in, 178 DMAIC process. See Define, measure, analyze,
massive open online courses (MOOCs), improve, and control (DMAIC)
186, 191 process
masters programs in, 186, 187t–190t DMS. See Document management systems
most related disciplines and fields for, (DMSs)
172–173 DNN. See Deep neural networks (DNN)
as new paradigm, 146–147 Document management systems (DMSs),
PhD programs in, 182–186, 183t–185t 78–81, 79f–81f
professions of doing, 173–174 Dogster, 200
purposes of analysis and research Domain knowledge, 19–20, 26
questions in, 175 Domo, 252
shift in practices and access, 144–146 Donoho, 182
wider concept of knowledge management Dredze, M., 215
and, 143–144, 172–173, 173f Driving, intangible assets and metrics in
word cloud and clusters on qualifications automated, 347–351, 348t
and responsibilities in, 178–180, DSMSs. See Data stream management systems
178t, 179f, 180t, 181f (DSMSs)
 Index ◾ 437

DSS. See Decision support systems (DSSs) F

Dunnhumby, 11t, 19
Durkheim, E., 199 Facebook, 79, 162, 174, 196, 198, 214
DW. See Data warehouse (DW) repositories analytics, 201, 203
Dyché, J., 9, 16 data acquisition by, 208
launch and growth of, 200–201
NodeXL and, 212
E Fairclough, N., 384
eBay, 11t, 17, 203 Fan, W., 204
Economist Intelligence Unit, 3, 5 Fawcett, T., 171, 174
Eding, R., 16 Fayyad, U. M., 32
Education, data science Feature scaling, 297
bootcamps, 191 Features extraction, 297–298
graduate certificate programs, 186 Feedback. See User feedback
integrated design for, 191–192 Ferguson, C. A., 384
massive open online courses (MOOCs), Financial services
186, 191 intangible assets and metrics in, 344–347,
masters programs, 186, 187t–190t 345t
PhD programs, 182–186, 183t–185t social media analytics in, 214
programs overview, 182–191 Finegan, E., 384
Eggins, S., 384 Fishwick, P. A., 283
eHarmony, 11t, 22 FiveThirtyEight, 158
Elbattah, M., 282 Flickr, 212
Elderly discharge planning simulation Ford Motor Company, 11t, 14
modeling, 291–292, 292t Forrester, J. W., 279
Ellison, N., 199 Fotolog, 200
e-mail, 199 Freeman, L., 212
customer contact via, 19, 21–22 Frey, C. B., 13
cyber threat intelligence and, 410 Friends+Me, 214
linguistic inquiry word count and, 425 Friendster, 200
Listserv and, 200 Future developments in analytics, 23–25
social media and, 159, 176, 200–201 Future Foundation/SAS, 24
time dimension, 92
Employers, data science, 177–178, 177f G
Ensembles
and bagging, 56 Galar, M., 310
and boosted trees, 56 Gandomi, A., 284
and information fusion, 56 Gartner Group, 10
Enterprise resource planning (ERP) Gartner Research, 6
systems, 74 Gartner’s 3Vs, 286
Epinions, 200, 205 GATE, 213
Equifax, 11t, 17–18, 22–23 Gee, J., 384
ERP. See Enterprise resource planning (ERP) Gelderse Vallei hospital, 11t, 16
systems Geneia, 11t, 18
European Court of Justice, 24 Genentech, 162
Evaluation and data visualization, 274 General Electric Network for Information
Everett, M., 212 Exchange (GEnie), 200
Everypost, 214 Geographical classification, automated, 368t,
Exchange analysis, 231 379t–380t, 392t–393t
Explicit knowledge, 333 Geographic information systems (GIS), 94–97
Explore step, SEMMA process, 41 Gilder, A., 215
438 ◾ Index

Glassdoor.com, 161 Hornik, K., 54

Goodwin, B., 15 Huang, Y., 290
Google, 200 Human intelligence (HUMINT), 415–416
Analytics 2.0, 201 Hybrid simulations, 282–283
Analytics 3.0, 203
Google Alerts, 213 I
Google Analytics, 155, 162
Google Earth, 157 IBM, 10, 152, 213–214
Google Maps, 157 4Vs, 286
Google Search, 213, 408 IC. See Intellectual capital (IC)
Google Trends, 213 Impact analysis, 231, 233t–234t
Gordon, M. D., 204 Improve step, DMAIC process, 45
Graduate certificate programs in data science, Indeed.com, 161
186 Infographics, 151
Greylock Partners, 174 Information visualization, 148–151, 151f
Gruber, T., 68, 77 Infrastructure and data visualization, 274
Instagram, 147, 198
H Institute for Operations Research and
the Management Sciences
Hadoop, 154–155, 180, 202, 213 (INFORMS), 6
Haenlein, M., 198 Intangible assets
Haider, M., 284 in automated driving, 347–351, 348t
Haight, J., 7 in financial services, 344–347, 345t
Halaburda, H., 22 in healthcare, 340–344, 341t
Halliday, M. A., 384 illustrative applications of metrics of,
Hall, P., 26 340–351
Halper, F., 21 scenarios, 336–340, 337t
Hamilton, M., 384 Intangible dynamics
Harper, P. R., 283 in automated driving, 347–351, 348t
Harris, J. G., 6, 10 Big Data and business analytics and
Hastie, T., 56 intelligence, 328–329
Hawamdeh, S., 172–173 conclusions, 351–352
Hawking, S., 25 in financial services, 344–347, 345t
Healthcare, intangible assets and metrics in, in healthcare, 340–344, 341t
340–344, 341t illustrative applications, 340–351
Heer, J., 174 intelligence hierarchy in organizations and,
Hellerstein, J. M., 174 333–335
Henke, N., 23 introduction to, 326
Henzl, V. M., 384 measuring intangible asset scenarios in,
HEQCO, 11t 336–340, 337t
Hey, T., 146 reimagining the intangibles hierarchy and,
Hierarchical data visualization, 260–265, 329–333
261f–265f wider view of, 326–328
Higher Education Quality Council of Ontario, Integration, data, 176
20 Intel, 348t
Hilton, N. A., 288 Intellectual capital (IC), 326–328. See also
Histograms, 255, 256f Intangible assets
Historical data, 207 Intelligence hierarchy in organizations,
Hive, 180 333–335
Holsapple, C., 9–10, 203–205 International Consortium of Investigative
Hong Kong Efficiency Unit, 11t, 14 Journalists (ICIJ), 20
Hootsuite, 214 International Women’s Day, 92–94, 98–101
 Index ◾ 439

Internet of Things (IoT), 7 KPI. See Key performance indicators (KPIs)

Internet relay chat (IRC), 200 KPMG, 162
IPSoft, 11t Kress, G., 384
Kurtz, C. F., 329–333
J
L
Japkowicz, N., 310
Java (programming language), 180 Ladder of aggression, 419
J. Lyons & Co., 9 Lane, D. C., 288
Jordan, M. I., 209 Laney, D., 284
Jubert, A., 70 Language. See also Text analysis
Jupyter, 252 knowledge structures represented in source
evidence, 385
K levels of, 382–384, 383f
linguistic inquiry word count (LIWC),
Kalantzis, M., 384 421–427, 426f
Kandel, S., 174 linguistic registers represented in source
Kaplan, A. M., 198 evidence, 384–385
Kaplan, D., 384 meaning or understanding derived from,
KDD. See Knowledge discovery in (KDD) 384
databases process reducing subjectivity and increasing
Keim, D. A., 252 objectivity in working with source
Key performance indicators (KPIs), 74, 328 evidence on, 386
Khan, G. F., 76 return on investment for analysis of, 387
KI. See Knowledge infrastructure (KI) semantic methods for, 386
Kirby, M. W., 8–9 text analysis in cyber threat intelligence,
KM. See Knowledge management (KM) 420–427, 426f
k-means algorithm, 298, 299t Language based analytics and qualitative
Knowledge analysis, 372–374
derived from user feedback, 130–137, Language Computer Corporation, 213
130f–132f, 135f Last.fm, 200
explicit, 333 Latent Dirichlet allocation (LDA), 98, 210
intelligence hierarchy in organizations, Latent Semantic Analysis (LSA), 98, 210
333–335 Lattner, A. D., 283
tacit, 78, 333 Layout in data visualization, 272–273
Knowledge discovery in (KDD) databases LDA. See Latent Dirichlet allocation (LDA)
process, 32–33, 33f Learning, 70
Knowledge infrastructure (KI), 71 Lee, L., 209
Knowledge management (KM), 5, 356–357 Lee-Post, A., 9–10
data science related to, 143–144, 172–173, Lerman, K., 215
173f Leskovec, J., 128
evolution of analytics in, 71–75, 72f–73f Lewandowsky, S., 271
research and business goals, 359–362, Lexalytics Sentiment Toolkit, 213
360t–370t Liang, Y. E., 204
user feedback and, 136–137 Line charts, 254–255, 255f
wider view of intangibles in, 326–328 Line graphs, 258–259
Knowledge Management Corporation (KMC), LingPipe, 213
253 Linguistic inquiry word count (LIWC),
Koehler, A., 92 421–423
Kosslyn, S., 272 case study, 425–427, 426f
Kotorov, R., 14 SentiStrength and, 424–425
Kourie, D., 132 Linguistic registers, 384–385
440 ◾ Index

Link analysis, 127–128 McKinsey & Company, 9, 173, 336

LinkedIn, 79, 161–162, 174, 200–201 McKinsey Global Institute, 3, 335
Listserv, 200 Measurement and business goals, 360–361
Lithium Social Media Monitoring and Measure step, DMAIC process, 44
Trackur, 213 Mehta, R. G., 297
Liu, B., 125 Metrics, 206, 335
LIWC. See Linguistic inquiry word count (LIWC) illustrative applications of intangible assets
Location dimension, social media analytics, and, 340–351
94–97 measuring intangible asset scenarios in,
Logistic regression, 55 336–340, 337t
Lopez-Rabson, T., 20 Mezei, J., 10
LSA. See Latent Semantic Analysis (LSA) MHR Analytics, 3, 10
LunarStorm, 200 Microsoft Corporation, 146
Luxottica Retail North America, 11t, 13 6Vs, 286
Lynda.com, 162 Microsoft Power BI, 214
Minard, C., 150, 151f
M Mixed methods analysis and variant sources,
374–375
Machine learning (ML). See also Simulation opportunities for, 377
modeling (SM) use cases, 377t–381t
artificial intelligence-assisted simulations Mixi, 200
and, 283 ML. See Machine Learning (ML)
conclusions on, 319 MLP. See Multilayer perceptron (MLP)
hybrid simulations, 282–283 Mobileye, 347
integration with simulation modeling, Model building, CRISP-DM, 37–38
280–281 Model step, SEMMA process, 41–42
modeling cluster-based flows of patients, Modify step, SEMMA process, 41
302–305, 303f, 303t–305t Modjeski, R. B., 283
prospective role in simulation modeling, Mohammad, S. M., 127
281–282, 281f Molloy, O., 282
sentiment analysis and, 209 Monitoring tools, social media, 213
supervised, 308–312, 310t, 311f, 313f–314f, MOOCs. See Massive open online courses
313t–314t (MOOCs)
unsupervised, 296–301, 297f, 299f–302f, Motley Fool, 205
299t Motorola, 42
Magruder, J., 215 MoveOn.org, 200
Managerial analytics, 7, 14–17 MUD. See Multi-User Dungeon (MUD)
Mao, H., 214 Multidimensional visualization, 253–260,
Map visualization, 262f–263f 254f–260f
Marketing, social media analytics in, 215 Multilayer perceptron (MLP), 54, 55f
Marks and Spencer (M&S), 11t, 15, 22 Multiple projects, 20–22, 21f
Marr, B., 24 Multi-User Dungeon (MUD), 199
Massive open online courses (MOOCs), 186, 191 Musk, E., 25
Masters programs in data science, 186, 187t–190t Mustafee, N., 282
Mathematica, 212 MySpace, 200
Mathematics language and knowledge
structures, 366t, 378t, 390t–391t N
MATLAB ®, 211–212
MavSocial, 214 Napster, 200
McCloy, U., 20 National Health System (NHS), 15
McDermott, R. A., 70 Natural language processing (NLP), 17, 152,
McIntosh, A., 384 162. See also Text analysis
 Index ◾ 441

language based analytics and, 372 P

sentiment analysis and, 208–209
text analysis tools, 213 Paepcke, A., 174
Navigant Research, 347–348 Pakath, R., 9–10
NDT. See Neural Decision Technology (NDT) Panama Papers, 11t, 20
NetDraw, 212 Pang, B., 209
Network visualization tools, 212 Parallel coordinates plots, 264–265, 265f
Neumann, S., 384 Park, H., 7
Neural Decision Technology (NDT), 17–18, 22 Patel, V. R., 297
Newell, A., 279 Patient clusters unsupervised machine
Ng, A. Y., 209 learning simulation, 296–301, 297f,
NHS. See National Health System (NHS) 299f–302f, 299t
Nightingale, F., 8, 12 Pauwels, K., 215
Nir, B., 384 Pereira, F., 215
NLP. See Natural language processing (NLP) Pfizer, 162
NOCA, 296 PhD programs in data science, 182–186,
NodeXL, 212 183t–185t
Nonaka, I., 69–70, 78, 326–327 Physician-patient communications, 369t, 380t,
Nucor Corporation, 4 393t–394t
nuTonomy, 348t Pie charts, 260
Pig (programming language), 180
Piskorski, M. J., 22
O Plotly, 252
Obama, B., 158 PMML. See Predictive Model Markup
Obar, J. A., 198 Language (PMML)
Observation of user actions, 123–124 Polanyi, M., 4
Ocado, 11t, 19, 21–22 Political environment, 24–25
O’Connell, R., 92 Political terror scale, 418, 419t
On-Road Integrated Optimization and Polling and politicians, 158–159, 206–207, 215
Navigation (ORION), 16, 22, Powell, J., 282
202–203 Power BI, 252
Operational analytics, 8, 17–18 Predictive analytics, 6–7
Operations research/management science, in cyber threat intelligence, 418–420, 419t
9–10 Predictive modeling, care outcomes, 308–312,
Opinion dimension, social media analytics, 310t, 311f, 313f–314f, 313t–314t
105–110 Predictive Model Markup Language (PMML), 25
Opinion-location interaction, social media Provost, F., 171, 174
analytics, 110–113 Prusak, L., 168
Opinion mining, 125–127 Public interest data, 169
Organizational environment, 23–24 PwC, 3, 5, 7
Organizations Python (programming language), 149, 154,
analyzing the value network of, 231–238, 162, 180, 212
233t–234t, 236t–237t text analysis tools, 213
intelligence hierarchy in, 333–335
patterns of interaction as analytical design Q
elements in, 226
as self-adapting complex systems, 224–225 Qiu, B., 129
tangible and intangible transactions in, Qlik, 252
226–227 Qlik Sense, 214
Twitter-based communities as, 77–81 Qualitative analysis, 371
value network representation of, fitting tools to methods and sources in,
227–231, 230f 396–397, 398t
442 ◾ Index

Qualitative analysis (Continued) S

language based analytics and, 372–374
sources analyzed using, 381–388, 383f, Saltelli, A., 57
388t–396t Sample, explore, modify, model, assess
use cases, 377t–381t (SEMMA) process, 39–42, 40f, 43f
when to use, 375 Sample step, SEMMA process, 41
Quantitative analysis, 371–372 Sam’s Club, 158
fitting tools to methods and sources in, SAS Enterprise Miner (SAS EM), 152–153,
396–397, 398t 176–177, 179–180, 212
sources analyzed using, 381–388, 383f, SAS Institute, 40
388t–396t SAS Sentiment Analysis Manager, 212
use cases, 377t–381t SAS Visual Analytics, 214
when to use, 375 Scatterplots, 257–258, 258f–259f
Quinlan, J., 54 Schiffman, H., 384
Schneider National, 11t, 15
Scientific analytics, 8, 20
R Scientific data, 168
R (programming language), 149, 154, Scientific programming tools, 211–212
178, 211 SD. See System dynamics (SD)
Rabelo, L., 282 SEB. See Skandinaviska Enskilda
RAND Corporation, 13 Banken (SEB)
Random forests algorithm, 312, 313f, 313t Second Life, 198
RapidMiner, 212 Self-adapting complex systems
Rashwan, W., 283–284 organizations as, 224–225
Ravid, D., 384 patterns of interaction as analytical design
RDBM. See Relational database management elements in, 226
(RDBM) SEMMA. See Sample, explore, modify, model,
Real-time data, 207 assess (SEMMA) process
Reinhold, O., 204 Sendible, 214
Relational database management Sensitivity analysis, 56–57
(RDBM), 74 Sentiment analysis, 125–127, 208–209
Repertory grids, 241–245, 243t in cyber threat intelligence, 423–424
Research and business goals, 359–362 SentiStrength, 424–425
analytical methods best suited to, Sentiment classification, 126–127
371–375, 371t Sentiment composition, 127
good practices and good practice SentiStrength, 424–425
models, 360 Shahani, A. K., 283
key questions by dimensions of the Shannon, R. E., 283, 288
framework, 358t Sharda, R., 75
level of investment, 361–362 Shirley Research Institute, 9
level of risk, 361 Silver, N., 158
measuring the results of analysis, Simon, H. A., 279
360–361 Simulation-based healthcare planning,
project or enterprise-level goals, 362 283–284
understanding why in context, 362, Simulation modeling (SM). See also Machine
363t–370t learning (ML)
Robinson, A., 7 agent-based modeling in, 288, 289f, 289t
Rowland, K., 14, 22 artificial intelligence-assisted simulations,
RSS feeds, 207–208 283
Ruggiero, A., 204 Big Data, analytics, and, 284–290,
Rutz, O. J., 215 285f–287f, 287t, 289f, 289t, 291f
 Index ◾ 443

cluster-based flows of patients, 302–305, Social media analytics (SMA), 75–81, 115–116,
303f, 303t–305t 196–199, 197f, 198t, 215–216. See
conclusions on, 319 also Twitter-based communities
data description, 293–296, 294f–296f (TBCs)
discovering patient clusters unsupervised business applications, 212
machine learning, 296–301, 297f, collective knowledge within communities of
299f–302f, 299t practice, 69–71
discrete event simulation (DES) in, 288, conclusion and road ahead, 113–115
289f, 289t, 305–308, 306f–308f in cyber threat intelligence, 415–418
elderly discharge planning analytics use data visualization tools, 213–214
case, 291–292, 292t defining, 203–204
future directions, 317–318, 318f evolution of analytics and, 201–203
hybrid simulations, 282–283 evolution of analytics in knowledge
integration with machine learning, 280–282 management and, 71–75, 72f–73f
introduction to, 279–280 interaction among users, 85–90, 91f, 91t
model verification and validation, introduction to, 68–69
316–317, 317f location dimension, 94–97
motivation, 280–282 network visualization tools, 212
overview of analytics approach, 293 opinion dimension, 105–110
patient’s care journey, 305–308, 306f–308f opinion-location interaction, 110–113
prospective role of machine learning in, processes of, 204–207
281–282, 281f representative fields of, 214–215
related work, 282–284 scientific programming tools, 211–212
results and discussion, 314–316, 315f–316f sentiment analysis, 208–209
simulation-based healthcare planning, social media management tools, 214
283–284 social media monitoring tools, 213
study limitations, 318–319 stream processing, 211
supervised machine learning, 308–312, techniques, 207–211
310t, 311f, 313f–314f, 313t–314t text analysis tools, 213
system dynamics (SD) in, 288, 289f, 289t, time dimension, 92–93, 94f
302, 303f topic dimension, 97–101
unsupervised machine learning, 296–301, topic modeling, 209–211
297f, 299f–302f, 299t topic-time interaction, 102–104, 104f
Single-project analytics projects, 12–20 user dimension, 81–85. See also User
SixDegrees.com, 200 feedback
Six Sigma, 42–45, 44f visual analytics, 210–211
Skandinaviska Enskilda Banken (SEB), 11t, 19, Social media data, 169
21–22 Social media management tools, 214
Sky blog, 200 Social media monitoring tools, 213
Slade, D., 384 Social networking sites (SNSs), 73, 79. See also
SM. See Simulation modeling (SM) Social media analytics (SMA)
SMA. See Social media analytics (SMA) analyzing user feedback on, 124–128
Smart cities, 157 growth of, 196–197
Smartphones, 145 historical perspective of, 199–201
Smith, T., 22 link analysis, 127–128
Snapchat, 79 opinion mining, 125–127
Snowden, D. J., 329–333 platforms, 198–199
Snow, J., 8, 12 polling and, 158–159
SNS. See Social networking sites (SNSs) user response in, 124
Snyder, W., 70 Social Pilot, 214
Snyman, R., 132 Sopra Steria, 2
444 ◾ Index

Source evidence, 381–388 T

case examples, 388t–396t
fitting tools to methods and, 396–397, 398t Tableau, 149, 152–154, 162, 214, 252
linguistic registers represented in, 384–385 Tacit knowledge, 78, 333
nature of language, information, and Takeuchi, H., 78, 326–327
knowledge used as, 382–384 Tangible and intangible transactions, 226–227
reducing subjectivity and increasing Target stores, 11t, 17
objectivity in working with, 386 TBC. See Twitter-based communities (TBCs)
semantic methods for, 386 Technical type analytics projects, 6–7
Spark, 154–155 10-fold cross validation, 312, 313t–314t
Spence, I., 271 Tesco, 11t, 12–13, 18–19, 21, 24
Spender, J. C., 4 Tessone, C. J., 87
Spigelman, J. J., 384 Testing and evaluation, CRISP-DM, 38–39
Sprout Social, 214 Text analysis, 356–357. See also Language
SPSS, 180, 212–213 case examples, 388t–396t
Srinivasan, S., 215 competencies needed to support, 387
Stakeholders. See Organizations conceptual framework, 357, 358t
Stascavage, J., 22 in cyber threat intelligence, 420–427, 426f
Statistical process control (SPC), 15 feasibility without semantic methods, 387
STATISTICA Text Miner, 213 level of effort involved in doing rigorous
Stephen, S., 310 analysis of, 387
Stieglitz, S., 206 linguistic inquiry word count (LIWC),
Stinchcombe, M., 54 421–427, 426f
Stodder, D., 174 return on investment for, 387
Storey, V. C., 9–10 SentiStrength, 424–425
Storytelling, 240 tools, 213
Strategic analytics, 7, 12–14 workflow, 386
Stream processing, 211 Text in data visualization, 271
Strevens, P., 384 Thangavel, K., 297
Student attrition, 46–57 ThirdVoice, 200
analytics methodology, 49–57, 51f, Time dimension, social media analytics, 92–93,
51t–53t, 55f 94f
data description, 50–54, 51f, 51t–53t Tinto, V., 48
discussion and conclusions, 59–63, Tobin’s q, 336, 338–339
60t–61t, 61f Tolk, A., 290
introduction and motivation, 47–49 Tonnies, F., 199
predictive analytics models, 54–56, 55f Topic dimension, social media analytics,
results, 57–59, 57t–59t 97–101
sensitivity analysis, 56–57 Topic modeling, 209–211
Subjectivity classification, 126 Topic-time interaction, social media analytics,
Süddeutsche Zeitung, 20 102–104, 104f
Sun, C., 130 Trafficking in persons semantic discovery, 370t,
Sun, Y., 310 381t, 394t–395t
Supervised machine learning, 308–312, 310t, Transactional value analytics. See Value
311f, 313f–314f, 313t–314t network analysis (VNA)
Support vector machines (SVMs), 55–56 Transaction data, 169
discussion and conclusions, 59–63, 60t–61t, Tree graphs, 263–264, 264f
61f Tree maps, 261, 261f
results, 57–59, 57t–59t Treleaven, P. C., 198, 214
Sweetser, A., 288 Tribe.net, 200
System dynamics (SD), 288, 289f, 289t, 302, 303f Trump, D., 158
cluster-based model, 303–305, 304t–305t TS-Prolog, 283
 Index ◾ 445

Tufte, E., 151, 267 deriving knowledge from, 130–137,

Tweet-level analysis, 134–135 130f–132f, 135f
Twitter existing work on analyzing, 128–130
data acquisition by, 208 introduction to, 121–123
data generated and transferred by, 147, 159 knowledge management and, 136–137
growth of, 196, 201 link analysis, 127–128
launch of, 200–201 opinion mining, 125–127
news obtained from, 199 User generated content (UCG), 196–198
NodeXL and, 212 User interactions, social media, 85–90, 91f, 91t
Search API, 211 U.S. presidential campaign, 2012, 364t–365t,
sentiment analysis, 208–209, 424 378t, 389t–390t
Twitter-based communities (TBCs), 68, 76. See
also Social media analytics (SMA); V
User feedback
as communities of practice, 76–77, 115–116 Value creation analysis, 231, 236t–237t
conclusion and road ahead, 113–115 Value network analysis (VNA)
interaction among users, 85–90, 91f, 91t analyzing the value network, 231–238,
knowledge represented in dimensional 233t–234t, 236t–237t
model, 78–81, 79f–81f conclusive summary, 246–247
location dimension, 94–97 developing commitment for organizational
opinion dimension, 105–110 move, 245–246
opinion-location interaction, 110–113 eliciting methodological knowledge,
as organizations, 77–81 238–241
time dimension, 92–93, 94f ensuring coherent value creation and, 238
topic dimension, 97–101 externalizing value systems through
topic-time interaction, 102–104, 104f repertory grids, 241–245, 243t
transforming tacit knowledge in, 78 introduction to, 222–223
user dimension, 81–85 organizations as self-adapting complex
systems and, 224–225
U patterns of interaction as analytical design
elements in, 226
Uber, 11t, 13, 348t, 350 purpose of, 223–224
UCG. See User generated content (UCG) tangible and intangible transactions in,
UCINET, 212 226–227
UIMA. See Apache Unstructured Information value network representation of
Management Applications (UIMA) organizations, 227–231, 230f
Ungson, G. R., 77 Vergne, M., 129
Unsupervised machine learning, 296–301, Verhoeven, L., 384
297f, 299f–302f, 299t Verizon, 162
UPS, 9, 11t, 15–16, 22, 202–203 Violin plots, 257, 257f
Uptake, 162 Virgo, P., 23
Ure, J., 384 Visalakshi, N. K., 297
U.S. Air Force, 11t, 13 Visual analytics, 210–211
Use case scenarios, 362, 363t–370t, 377t–381t Visual Display of Quantitative Information,
Usenet, 199 The, 267
User dimension in Twitter-based communities, Visualization, data
81–85 color in, 271–272
User feedback hierarchical, 260–265, 261f–265f
analyzing, 124–128 implications for future directions, 273–274
collecting, 123–124 information, 148–151, 151f
conclusions and future work, 137–138 introduction to, 251–253
data management, 133–134 layout in, 272–273
446 ◾ Index

Visualization, data (Continued) White, H., 54

multidimensional, 253–260, 254f–260f Whole Earth Lectronic Link (WELL), 200
network, 212 Wikipedia, 198, 200
practice, 253–265, 254f–265f data acquisition by, 208
principles of, 265–273, 268t–270t Wildman, S. S., 198
text in, 271 Wittwer, M., 204
tools, 213–214 Wolfram, M. S. A., 214
VNA. See Value network analysis (VNA) WordStat, 213
Voice of the customer (VoC), 18 World of Warcraft, 198
Vos, M., 204
Y
W
Yahoo!, 200
Wainer, H., 267 Yang, J., 128
Walsh, J. P., 77 Yang, Z., 87
Ward, J. S., 284 Yau, N., 266
Warehouse Group, 11t, 14, 22 Year 2000 problems, 12
Warehouse Stationery, 14 Yelp, 215
Watson Analytics, 152, 214 YouTube, 196, 198, 200
Waymo, 348t NodeXL and, 212
Web data, 169
Web 2.0 tools, 71, 124 Z
WECO, 156
Weka ML, 212 Zeng, X., 214
WELL. See Whole Earth Lectronic Link (WELL) Zhong, J., 282–283
Wenger, E., 70, 77 Zika virus, 82–85, 88–91, 95–97
WhatsApp, 196 Zuckerberg, M., 200