Data-Driven Marketing

Session 4 – Text and Data Mining + Visual Mining

Session
Session44
1. Preparation & Discussion
2. Text Mining
3. Data Mining
4. Social Listening & Sentiment
Analysis
5. Visual Mining
6. Group Task
From: Berger, Humphreys, Ludwig, Moe, Metzer &
Schweidel (2019)

A wealth of textual data

7
From: Colicev, Malshe, Pauwels, and O’Connor 2018

The nature and importance of customer mind-set

metrics
 Nature: They capture the “hearts and minds” of consumers
in different steps of the consumer decision journey: brand
awareness, purchase intent and post-purchase satisfaction
 Importance: They reflect the consumer reactions to social
media chatter and have an effect on firm’s stock market
value. Thus, they represent the intermediate step between
marketing actions and firm performance.
From: Berger, Humphreys, Ludwig, Moe, Metzer & Schweidel (2019)

The value of text to managers

• It is often more accurate as it occurs naturally in marketplace actions
• Is often archived, and therefore available to study trends over time
• Is freely available for both the focal brand and competitors
• Can access otherwise hard-to-reach consumer opinions, but must be aware of selection bias
and mindful of careful sampling
From:

Particularly earned media is often comprised of

textual data

10
From: Colicev, Malshe, Pauwels, and O’Connor 2018

Study and Overall Findings

Studied 45 brands over 21 sectors by combining 5 daily-level datasets
Consumer Mind-Set Metrics:

Earned Social Media Purchase Intent

(ESM): Volume of ESM improves
User-created social purchase intent: consumers
media posts trust other consumers more
when making purchase
decisions
Stock Market
Brand Awareness
Performance
Volume of ESM improves
brand awareness: frequent
Owned Social Media ESM creates more exposure
(OSM): to brand identifiers
Brand-controlled social
media posts Customer Satisfaction
Valence of ESM improves
satisfaction: consumers use
positive and negative ESM as
diagnostic information for
post-purchase satisfaction
From: Berger, Humphreys, Ludwig, Moe, Metzer & Schweidel
(2019)

Example
Brand positioning
map produced from
message board
discussion of car
brands based on
co-occurrence of
brands within each
post
868,174 consumer messages
consisting of nearly 6 million
sentences posted by 76,587
unique consumers between the
years 2001 and 2007.
Netzer, O., Feldman, R., Goldenberg,
J. and Fresko, M., 2012. Mine your
own business: Market-structure
surveillance through text
mining. Marketing Science, 31(3),
pp.521-543.
Video
Text and
Data Mining
(TDM)

AirFrance

16
TDM Algorithms

What is the Algorithm(s) behind TDM?

How does it work?

Who would execute it?

When would it be used?

Limitations thereof?
17
Definitions
• Text Mining - A branch of analytics that encompasses the treatment of unstructured text information. Automatic grouping of similar texts is
a task of great interest in various economic and social domains.

• Data Mining - The exploration and analysis of data in order to uncover patterns or rules that are meaningful
• Predictive Analytics - Predictive analytics is a branch of advanced analytics that makes predictions about future outcomes using
historical data combined with statistical modeling, data mining techniques and machine learning

• Social Listening - Process of identifying/assessing what is being said about a company, individual, product or brand on the internet
• Sentiment Analysis - Sentiment analysis is the process of analyzing online text to determine the emotional tone they carry. It aims
to detect whether sentiment around a brand or topic is positive, negative, or neutral. Simply put, sentiment analysis determines how the author
feels about a certain topic.

• Visual Mining - Extracting trends, patterns, insights and knowledge from visual data with the help of data mining, computer vision and
data visualization techniques.

18
Session
Session44
1. Preparation & Discussion
2. Text Mining
3. Data Mining
4. Social Listening & Sentiment
Analysis
5. Visual Mining
6. Group Task
Definitions
• Text Mining - A branch of analytics that encompasses the treatment of unstructured text information. Automatic grouping of similar texts is
a task of great interest in various economic and social domains.

• Data Mining - The exploration and analysis of data in order to uncover patterns or rules that are meaningful
• Predictive Analytics - Predictive analytics is a branch of advanced analytics that makes predictions about future outcomes using
historical data combined with statistical modeling, data mining techniques and machine learning

• Social Listening - Process of identifying/assessing what is being said about a company, individual, product or brand on the internet
• Sentiment Analysis - Sentiment analysis is the process of analyzing online text to determine the emotional tone they carry. It aims to
detect whether sentiment around a brand or topic is positive, negative, or neutral. Simply put, sentiment analysis determines how the author
feels about a certain topic.

• Visual Mining - Extracting trends, patterns, insights and knowledge from visual data with the help of data mining, computer vision and
data visualization techniques.

20
From: Berger, Humphreys, Ludwig, Moe, Metzer & Schweidel
(2019)

How automated text analysis works

• A computer-assisted methodology to describe “what” is

being said and “how” it is said.
• Considers individual words and expressions, their linguistic
relationships within a document and across documents as
well as the more general topics discussed in the text.
• Techniques range from computerized word-counting and
applying dictionaries to supervised or automated
machine learning that help deduce meanings of text.
From:
 Tokenization • Text divided into smaller characters called
Algorithm tokens
Example Elimination of stop
• Words with no meaning removed (e.g. ‘and’,
words
‘the’)
Text Mining Stemming /
Lemmatization • Words reduce to their root or base form

Term Frequency-Inverse • Statistical calculation to determine the

Document Frequency (TF-IDF)
frequency and weight of a word/term

Named Entity • Identify and classify named entities within text

Recognition (NER)
(names, locations, etc)

Opinion mining / • Classifying text as positive, negative, or neutral to

Sentiment analysis
determine emotion

Topic Modeling & • Identify and extract hidden topics or themes &
Text Classification
assign predetermined labels or categories to text
23
based on text content
Text Mining Success

● Text Mining success depends on numerous factors:

○ Quality of the data (Garbage in, Garbage out)
○ Data cleansing
○ Algorithm selection
○ Scalability
○ Selection bias
○ Model evaluation
○ Ethical boundaries
24
28
Exmple: The Science of Happiness –Data profile

29
The Science of Happiness –Data profile

30
Preview: tidy and happy data

31
What were the most frequent elements of happiness?

32
33
Word associations (1/2)

34
Word associations (2/2)

35
Session
Session44
1. Preparation & Discussion
2. Text Mining
3. Data Mining
4. Predictive Analytics
5. Social Listening & Sentiment
Analysis
6. Visual Mining
7. Group Task
Definitions
• Text Mining - A branch of analytics that encompasses the treatment of unstructured text information. Automatic grouping of similar texts is
a task of great interest in various economic and social domains.

• Data Mining - The exploration and analysis of data in order to uncover patterns or rules that are meaningful
• Predictive Analytics - Predictive analytics is a branch of advanced analytics that makes predictions about future outcomes using
historical data combined with statistical modeling, data mining techniques and machine learning

• Social Listening - Process of identifying/assessing what is being said about a company, individual, product or brand on the internet
• Sentiment Analysis - Sentiment analysis is the process of analyzing online text to determine the emotional tone they carry. It aims to
detect whether sentiment around a brand or topic is positive, negative, or neutral. Simply put, sentiment analysis determines how the author
feels about a certain topic.

• Visual Mining - Extracting trends, patterns, insights and knowledge from visual data with the help of data mining, computer vision and
data visualization techniques.

37
Data Mining
The process

38
Algorithms
Association Rule • Discover interesting relationships or associations
Mining
Data Mining among items in a dataset

Examples Classification • Predict categorical labels or classes for new

instances based on training data
Neural Networks • Powerful models inspired by the structure of the
human brain
Clustering • Group similar data points together based on
their characteristics or attributes
Sequential pattern
mining • Discover patterns or relationships in sequential
data, such as time series or transactional data
Regression • Predict continuous numerical values. e.g. Linear
regression, logistic regression
Anomaly Detection • Identify unusual or anomalous patterns in
data. e.g. detecting fraudulent transactions 39
 example

Predicting Churn

Source: freecontent.manning.com 40
Session
Session44
1. Preparation & Discussion
2. Text Mining
3. Data Mining
4. Predictive Analytics
5. Social Listening & Sentiment
Analysis
6. Visual Mining
7. Group Task
Definitions
• Text Mining - A branch of analytics that encompasses the treatment of unstructured text information. Automatic grouping of similar texts is
a task of great interest in various economic and social domains.

• Data Mining - The exploration and analysis of data in order to uncover patterns or rules that are meaningful
• Predictive Analytics - Predictive analytics is a branch of advanced analytics that makes predictions about future outcomes using
historical data combined with statistical modeling, data mining techniques and machine learning

• Social Listening - Process of identifying/assessing what is being said about a company, individual, product or brand on the internet
• Sentiment Analysis - Sentiment analysis is the process of analyzing online text to determine the emotional tone they carry. It aims to
detect whether sentiment around a brand or topic is positive, negative, or neutral. Simply put, sentiment analysis determines how the author
feels about a certain topic.

• Visual Mining - Extracting trends, patterns, insights and knowledge from visual data with the help of data mining, computer vision and
data visualization techniques.

48
Sentiment analysis

49
Sentiment analysis in R can be based on a list of predefined
words

50
Social Listening
Social Listening Process
● Process of identifying
and assessing what
1. Set Marketing Objectives: In house
is being said about a 2. Data Collection: Tools/Platforms
company, individual, 3. Data Analysis: Tools/Platforms
product or brand on 4. Insights Generation: Tools/Platforms
the internet 5. Engagement and Response: In house
6. Monitoring: Tools/Platforms

51
Sentiment Analysis - Applications

● Brand reputation management and monitoring

● Customer research and feedback
● Market Research for decision making
● Crisis prevention or management
● Social media, Financial, Political analysis
● Customer service and support

52
Sentiment Analysis
Example

● Positive sentiment can be

expressed using words such as:
“good”, “great”, “wonderful”,
“happy” and “fantastic”

● Neutral sentiment can be

explained as lack of emotion
and bias, remaining objective

● Negative sentiment can be

expressed using words such as:
“bad”, “terrible”, “hate” and
“disgusting”

53
Sentiment Analysis - Boeing

● The company suffered heavy

reputational damage after the
crash of a Boeing 737-800 in the
China Eastern Airlines
● Through the use of sentiment
analysis and reputation score
this can be monitored

54
Session
Session44
1. Preparation & Discussion
2. Text Mining
3. Data Mining
4. Predictive Analytics
5. Social Listening & Sentiment
Analysis
6. Visual Mining
7. Group Task
Definitions
• Text Mining - A branch of analytics that encompasses the treatment of unstructured text information. Automatic grouping of similar texts is
a task of great interest in various economic and social domains.

• Data Mining - The exploration and analysis of data in order to uncover patterns or rules that are meaningful
• Predictive Analytics - Predictive analytics is a branch of advanced analytics that makes predictions about future outcomes using
historical data combined with statistical modeling, data mining techniques and machine learning

• Social Listening - Process of identifying/assessing what is being said about a company, individual, product or brand on the internet
• Sentiment Analysis - Sentiment analysis is the process of analyzing online text to determine the emotional tone they carry. It aims to
detect whether sentiment around a brand or topic is positive, negative, or neutral. Simply put, sentiment analysis determines how the author
feels about a certain topic.

• Visual Mining - Extracting trends, patterns, insights and knowledge from visual data with the help of data mining, computer vision and
data visualization techniques.

57
Visual data mining
Dash Hudson
Dash Hudson's cutting-edge solution combines the
power of our proprietary technology, Vision AI, with
comprehensive social listening insights integrated
throughout the entire platform.

Powered by Dash Hudson's proprietary Vision AI

technology, which has been continuously learning since
2016, the tool goes beyond traditional features like
monitoring topics, hashtags and keywords to surface
visual trends

With this data in hand, marketing teams can gain deeper

insight into community sentiment across social and
make informed decisions to drive success- CanadianInsider
April 2023
58
From: Tellis, MacInnis, Tirunillai, and Zhang (2019)
How to Make Your
 This Research
Video Ad Go Viral
- Three independent coders rated sample of 860 video ads
on over 60 characteristics of ad content
Bud Puppy Love
Examined impact on virality
Most Viral Ad in Sample
-

Results:
Use emotion more than information to engage viewer
- Current practice: 55% use information, few 15% use
strong emotion
 Use cute stimuli, babies/animals (rather than celebrities) for
authenticity
- Current practice: Only 3% use babies/animals, 26% use
celebrities
 Place brand name intermittently or at the end of the video to
avoid zapping
- Current practice: Only 1/3 do
Visual data
mining
Dash Hudson

Visual Mining

Rapidly changing landscape

60
Session
Session44
1. Preparation & Discussion
2. Text Mining
3. Data Mining
4. Predictive Analytics
5. Social Listening & Sentiment
Analysis
6. Visual Mining
7. Group Task
Group Task
Brand24
Choose a popular hashtag or marketing campaign
and monitor social media accounts (company’s and
followers) to gather insight on the general consumer
sentiment associated with the hashtag/ campaign.
(recent event that involved a relatively known
brand)

You can use social media monitoring platforms

available for free (trial version, e.g
https://app.brand24.com/) and monitor your topic over
a specific period of time (e.g. 30 days) to gather fresh
insights.
66
Group Task
Brand24

Questions
• What major event happened in the last week?
• What was the reason for this event
• Was there a response from the brand?
• What was it, would you have done the same or acted differently?
• What happened to their brand sentiment
• With hindsight, what would you as a Marketing Director suggest they have done or do with the
data?

67
Data-Driven Marketing

Master in Management