TWITTER DATA ANALYSIS USING SPARK

A Project

Presented to the faculty of the Department of Computer Science

California State University, Sacramento

Submitted in partial satisfaction of

the requirements for the degree of

MASTER OF SCIENCE

in

Computer Science

by

Yash Bopardikar

FALL
2016
 © 2016

Yash Bopardikar

ALL RIGHTS RESERVED

ii
 TWITTER DATA ANALYSIS USING SPARK

A Project

by

Yash Bopardikar

Approved by:

__________________________________, Committee Chair

Dr. Jun Dai

__________________________________, Second Reader

Dr. Ying Jin

____________________________
Date

iii
Student: Yash Bopardikar

I certify that this student has met the requirements for format contained in the University

format manual, and that this project is suitable for shelving in the Library and credit is to

be awarded for the project.

__________________________, Graduate Coordinator ___________________

Dr. Ying Jin Date

Department of Computer Science

iv
 Abstract

of

TWITTER DATA ANALYSIS USING SPARK

by

Yash Bopardikar

The idea is to solve people’s dilemma for choosing a certain product or service over

other. The objective is to collect data from tweeter feeds on trending topics like various

OS, new technologies, different Products etc. and to categories them according to need.

This Project would use Scala and Spark cluster. All the Big Data that would be fetched

would be collected in MySQL database. Then using the MapReduce API’s and Data

Mining algorithms data will be classified accordingly. Here we would be using Naïve

Bayes Algorithm for Sentimental analysis over the data. Twitter provides a streaming

API to stream real time twitter data. This library named Twitter4j is available in python

to access Streaming API and download twitter data. The filter on this data is based on a

list of keywords supplied. This analysis will employee a distributed data processing

system known as Apache Spark using several worker and master nodes. This cluster is

scalable and can handle millions of records.

To filter out the huge number of data we will use Map reduce technique on Spark. The

input file will contain a JSON object for each tweet in data. This file will be uploaded on

Spark frame structure. The Spark frame structure will have replicated the structure and

distribute to multiple nodes, thus the mapper takes all files presented in the directory and

v
classifies them according to the filter set. These filtered tweets will pass through the data

mining algorithms and thus we could observe one to one comparison of data which would

be helpful to take tough decisions.

_______________________, Committee Chair

Dr. Jun Dai

_______________________
Date

vi
 ACKNOLEDGEMENTS

I would like to thank Dr. Jun Dai, my advisor for providing me an opportunity to work on

this project and also motivate me to do better and better. His constant encouragement and

interest in making best out of such a powerful platform helped me throughout the project.

In addition, I would also like to thank Dr. Ying Jin for her willingness to serve on the

committee. Last but not the least, I would like to thank the entire faculty and staff of

Department of Computer Science Engineering at California State University,

Sacramento.

vii
 TABLE OF CONTENTS
Page
Acknowledgements ........................................................................................................... vii

List of Figures ..................................................................................................................... x

Chapters

1. INTRODUCTION .......................................................................................................... 1

1.1 Overview ................................................................................................................. 1

1.2 Introduction to Big Data ......................................................................................... 2

1.3 Sources of Big Data ...................................................................................................... 4

1.4 Why we need Big Data? ............................................................................................... 6

1.5 What is Big Data Analytics? ......................................................................................... 6

2. OVERVIEW TECHNOLOGIES AND PLATFORM USED ........................................ 7

2.1 Twitter ........................................................................................................................... 7

2.2 Apache Spark ................................................................................................................ 8

2.2.1 Spark Streaming ....................................................................................................... 11

2.2.2 Spark SQL................................................................................................................ 11

2.2.3 Spark ML Lib (Machine learning libraries) ............................................................. 12

2.3 SBT ............................................................................................................................. 12

2.4 Tableau ........................................................................................................................ 12

3. SYSTEM DESIGN ....................................................................................................... 13

3.1 Spark Cluster Setup..................................................................................................... 13

3.2 System Flow Diagram................................................................................................. 16

viii
4. IMPLEMENTATION ................................................................................................... 17

4.1 Streaming Twitter Data Using Spark .......................................................................... 17

4.2 Understanding Data .................................................................................................... 18

4.3 Pre- Processing Data ................................................................................................... 21

4.4 Developed Model ........................................................................................................ 24

4.5 Saving to Database...................................................................................................... 26

5. EXECUTION................................................................................................................ 28

5.1 Building Project .......................................................................................................... 28

5.2 Submitting Job ............................................................................................................ 29

5.3 Tableau ........................................................................................................................ 31

6. USE-CASES AND RESULTS ..................................................................................... 32

6.1 IPhone vs Samsung Note7 .......................................................................................... 32

6.2 PlayStation vs Xbox vs Nintendo ............................................................................... 37

7. FUTURE WORK .......................................................................................................... 40

8. CONCLUSION ............................................................................................................. 41

Bibliography ..................................................................................................................... 42

ix
 LIST OF FIGURES

Figure Page

1. V’s of Big Data ............................................................................................................... 2

2. Spark Architecture .......................................................................................................... 9

3. Spark Streaming ............................................................................................................ 11

4. Spark Master WebUI .................................................................................................... 14

5. Spark Worker UI ........................................................................................................... 15

6. Spark Application Window........................................................................................... 15

7. System Flow Diagram................................................................................................... 16

8. Spark Streaming ............................................................................................................ 17

9. DataFrame Schema ....................................................................................................... 21

10. Basic Workflow Model ............................................................................................... 24

11. DataFrame ................................................................................................................... 26

12. Tableau Connection .................................................................................................... 31

13. Density, Sentiment and Location Graph for Iphone vs Samsung .............................. 33

14. Density, Sentiment and Location Graph for Iphone vs Samsung .............................. 34

15. Popularity Graph Iphone v/s Samsung Note7............................................................. 35

16. Location Based Sentimental Analysis ........................................................................ 36

17. Heat Map for Gaming Consoles (Canada) .................................................................. 38

18. Heat Map for Gaming Consoles (India)...................................................................... 39

x
 1

Chapter 1

INTRODUCTION

1.1 Overview

In today's world there is a lot of informative data available in the social media. This data

has been a powerful resource for data miners to get deep insights. Collecting this data

and analyzing it might give us a lot of useful information which can help solve a lot of

our day to day problems. Taking this idea into consideration, I have harnessed a tool

which collects data from Twitter and analyses it to produce intelligent information.

Suppose we use this tool according to a consumer/buyer perspective, it can be used in the

following way. This app collects the Tweets that relate to brands, products or topics set as

filter and gives us a review as to which is the best for a given geographical location, what

are the sentiments associated with it at a particular location, which brand, product or topic

is most trending, what the trend looks like on different dates, how popular it is and how

many followers it has and thus helping the consumer to make wise choices. Consider

buying a gaming console as our target mission, we want to analyze gaming consoles that

belong to three brands say Xbox, PlayStation, Nintendo. This tool is not just limited to

gaming consoles but can be used to get reviews and insights for any product or any

trending topic for example phones, cars, electronic devices. The results can be visualized

it various graphs like heat map, bar graph, trending graph, density graph, etc.
 2
1.2 Introduction to Big Data

Today, we live in the digital world. Because of the growing digitization and innovations

in technologies, the amount of data (structured or unstructured) and data collection have

also tremendously increased. The data is deposited in databases which grow substantially

and become difficult to collect, manage, share, examine and visualize via traditional

database software tools [1]. This is the reason for emergence of big data and a recent area

of strategic investment for IT organizations.

Big Data can be defined as a collection of data which may be structured or unstructured

and so large and complicated that it may become complex to process it using simple

systems or traditional data processing applications [2]

Figure 1 V’s of Big Data [3]

 3

The characteristics of Big Data are the following [3]:

Ø Volume

Volume [3] refers to huge amount of data transformed every second which is not possible

for traditional systems to manage. For example, consider a social networking website like

Facebook where people upload pictures, send messages, comments and click the Like

button billion times on daily basis.

Ø Variety

The data like structured or unstructured, videos, audios, images, text messages, tables,

graphs etc., are referred to as Variety. These different forms of data can be collected via

various sources like social media, sensors, mobile devices, government datasets, user

based etc. This data can have many dimensions and also various forms.

Ø Velocity

The speed to generate new data and move around are stated as Velocity. For example, the

messages or videos on social media that’s going viral within a minute and quickly detect

the suspicious activity on credit card transactions. Systems designed to handle big data

can handle and process millions of rows per second, which makes it easier to get desired

insights out of this data.

Ø Veracity

Veracity refers to the reliability or disorderliness of the data [3]. The quality and accuracy

are not easy to manage with many forms of big data. For example, the post on Twitter or

Facebook with many hashtags, typos, and informal speech. The system has many worker
 4
nodes and threads running to complete a desired task which work in synchronization to

maintain the reliability of data.

Ø Value

Value refers to the business cost which is obtained through managing big data. For

example, managing very large unstructured data from blogs or social media stream. The

value of data depends on how accurate are the insights obtained from the data, the

processing time, how deep the mining is done and how it is visualized.

1.3 Sources of Big Data

With the evolution and innovation of technology, the amount of generated data is also

increased. Also, the data needs to be collected into variety of formats. Sources of Big

Data can be generally classified into six different categories:

Ø Enterprise Data

There is humongous amount of data spread across the various businesses, companies, and

institutions in different formats known as Enterprise Data. The formats include emails,

word documents, spreadsheets, presentations, HTML pages, pdf files, XMLs, flat files,

legacy formats, Json, csv, tables, graphs, etc.

Ø Transactional Data

There are countless applications like web applications, mobile applications, CRM

systems, banking systems, service providing platforms, etc., in every organization which
 5
involve in different kinds of transactions. So, the relational databases can be used as a

backend to reinforce these applications.

Ø Social Media

Social networks like Facebook, Twitter, etc., produce vast amount of data every day

which are unstructured data which can be text, images, video, and others. This data can

be cleansed, ordered and transformed into a structured data to make it useful data.

Ø Activity Generated

The data being generated by machines are referred to as Activity generated data. The

origin of these kinds of data comes from satellites, medical devices, sensor data,

industrial machinery, surveillance videos, cell phone towers, etc. This data may be

present in Log files, log tables, Json format and can be used to get future predictions of

the system.

Ø Public Data

The publicly available data like Wikipedia, data from weather department, data published

by research institutes, open data-sets provided by government and other types of data

which are easily available and accessible to public at no cost. This type of data is called

Public Data.

Ø Archives

In today's world, all insignificant data are being achieved by enterprises. As hardware

getting cheaper day-by-day, the enterprises can pile up the data. This kind of data

includes scanned documents and agreements, records of ex-employees, project

documents, and all banking records.

 6

1.4 Why we need Big Data?

A normal system hardware and software are not capable to deal with very large amount

of various types of data which are created and collected at such a high speed. Big data is

the term for large and complicated data sets that it becomes difficult for traditional data

warehousing to store, analyze, manage, process and work on them and visualize. The

insights gained by processing big data can be significant for a business, can help

consumers, can be used to get predictive models for natural calamities and avoid them,

and can be used to predict behavioral patterns and trends. That is the reason of big data

and its analysis is the main focus modern science and business.

1.5 What is Big Data Analytics?

“Big data analytics is the process of examining large data sets containing a variety of data

types, like big data, to uncover hidden patterns, unknown correlations, market trends,

customer preferences and other useful business information.” [4]

Using Big Data Analysis important insights can be gained. For this project we use Big

Data Analytic for getting insights from twitter data.

 7
Chapter 2

OVERVIEW TECHNOLOGIES AND PLATFORM USED

2.1 Twitter

Twitter [5] is a social networking platform used to share small bits of information across

the world. It is the most popular micro blogging website in today’s world. The 140

characters’ messages/posts are called tweet and people can follow each other to receive

other person's tweets. They can tag other using @ symbol.

Like all social media forum these tweets can include URLs, photographs, and hashtags. A

person can share another person's tweet (re-tweet) if he is following the later. Hashtag (#)

is used as a trendsetter in twitter, and used to easily look for information on a particular

topic while collecting data from twitter.

Twitter has more than billion registered user accounts and around 317 million monthly

active Twitter users [5]. It contains massive amount of data and included users from all

fields like movie stars, brand reviews, news, sportsmen, common people, politicians, etc.

thus giving us the global outlook of people around the world. Thus collecting the

expressions of people from different walks of life, can give us a clear picture on a

particular topic.
 8
2.2 Apache Spark

Spark [6] is a frame work which provides cluster computing platform to perform various

tasks like data analytics, machine learning, data streaming, database management,

parallel computing, graph operations, etc. Spark can run as a standalone cluster as well as

in cluster mode.

Spark uses RDD (resilient distributed dataset) to distribute items over the cluster. RDD’s

are read only datasets and are handled by spark for the purpose of maintaining its fault

tolerant behavior. Aside from RDD’s spark also has DataFrames which are table like

structures used by spark to store data in table format, for manipulation of data in

dataframes spark also provides SQL libraries and functions. Spark has Spark SQL library

which can be used to perform SQL queries on dataframes.

Apache Spark supports Java, Scala, Python and R [6]. Spark core is the main engine

which manages input output operations, scheduling, memory management, networking

interfaces and dataset (RDD based).

 9

The spark structure can be understood from the diagram below:

Figure 2: Spark Architecture [7]

Ø Driver Program

Driver program [7] manages the allocation of tasks to the worker nodes. It is also called

as Master. The driver does the task of listening to worker nodes for possible incoming

messages. It also keeps the task isolated so there is not data leak between the tasks. It

prioritizes the task of managing the jobs submitted to the spark cluster.

Ø Cluster Manager

Cluster manager [7] distributes the task to various worker nodes. It is like the

intermediator between the driver program and worker nodes which manages the cluster

when it is distributed. The cluster manager handles the request when either od driver or
 10
worker node is busy. This makes spark capable of fault tolerant. The cluster manager

supports various applications and package handlers

Ø Worker Node

Each worker node [7] has an executor which can perform many tasks. Every worker has a

cache memory allocated which is configurable. Every executor isolates the task so there

is no memory leak between the multiple tasks submitted. The only way share memory

between two tasks is to write it to an external memory.

Ø Executor

Executor [7] is the process initiated for an execution of an application. Each application

running on the cluster has its own executors. The executor is responsible for data keeping

the data and doing input output operations between the applications.

Ø Task

Task [7] is a unit of work assigned by an executor.

A spark cluster can be configured with various parameter settings. The number of

executors, worker nodes, memory allocation, cache memory, worker cores, executor

cores everything can be configured according to the system requirements.

Apache spark provides various other functionalities like:

 11
2.2.1 Spark Streaming

Spark Streaming is an add on package to the core Spark API which is scalable, high-

throughput and also fault-tolerant. It provides functionality of processing live data

streams. [8]. For streaming Apache spark can have flume, HDFS, apache Kafka, twitter,

kinesis data sources. This data can be then cleaned and structured in spark itself and used

to do further processing.

Figure 3: Spark Streaming [8]

2.2.2 Spark SQL

Spark core has an SQL extension which supports more optimization on datasets(RDD)

and is in structured format to retrieve data using the SQL queries. Spark SQL provides

most convenient way to perform several transitions on the data. Spark SQL [9] uses

dataframes for data manipulations.

 12
2.2.3 Spark ML Lib (Machine learning libraries)

Spark ML Lib [10] provides wide range of advanced machine learning libraries. Spark

has two kinds of libraries which perform operations on RDDs and DataFrames. Spark can

process large amount of data and perform advanced machine learning algorithms on it.

Spark supports clustering, classification, reduction, regression, etc.

2.3 SBT

“SBT is an open source build tool for Scala and Java projects.” [11]. It is tool for

compiling Scala code and integrating various Scala frameworks. SBT builds a jar file of

the project which includes all the dependencies needed for project to run. This jar then

can be deployed on various frameworks as a complete application with dependencies.

2.4 Tableau

Tableau [12] is data visualization software. Data visualization is one of the important

features of Big Data. Visualization views can project data in different dimensions which

can make an impact. Tableau is a powerful tool which can project data in many formats

and graphs. It has the power to extract data from various data sources as well as stream

data from data sources to provide a live view. Tableau has custom settings for color, toot-

tips, text representation, scale, etc. It has various versions which can be accessed online

as well as of-line as a desktop version.

 13
Chapter3

SYSTEM DESIGN

3.1 Spark Cluster Setup

I have set up spark cluster in standalone mode on my laptop which has 8GB

memory(RAM), 512GB storage and intel i5 processor. This spark cluster is build using

maven which is included in the spark package. Below are the spark environment

variables set for this cluster.

Here I have set spark worker instances to 4, which initiates 4 worker instances are soon

as the spark cluster is started. The Executor instances are set to 5 and each worker is

allocated 8 GB of memory i.e. the spark cluster can use the total amount of memory

available on the system. The spark Executor and Worker cores are set to 6 and 10 which

are virtual cores.

 14
Spark has a WebUI interface which provides visualizations for the processing and

resources allocated. I have set the WebUI port to 8080 on localhost.

The WebUI looks as follows for the system configured.

Figure 4: Spark Master WebUI

 15

Figure 5: Spark Worker UI

Figure 6: Spark Application Window

 16

3.2 System Flow Diagram

Figure 7: System Flow Diagram

 17
Chapter 4

IMPLEMENTATION

4.1 Streaming Twitter Data Using Spark

We use spark streaming to stream live spark data. To load Twitter data into Apache Spark

twitter provides an interface to developers which can be used to access twitter data. Visit

twitters applications site to register your application “https://apps.twitter.com/” [13]. I

have noted twitter tokens into TwitterKeys.txt which are needed to initialize spark

streaming context.

The streaming data is captured into batches. The interval for downloading batches can be

set, I have set the interval to be 10 seconds. Therefore, every 10 seconds a new batch of

data is streamed and captured into batches.

We set the filter which contains set of keywords to filter the tweets. Only tweets with

those keywords present in it would be streamed.

Figure 8: Spark Streaming [8]

 18

4.2 Understanding Data

The far most and initial stage in data mining is to perceive and recognize the problem and

to note down the objectives [14]. We have equipped ourselves with domain knowledge to

recognize the problem complication, it will tremendously improve data mining

effectiveness and potency. We have thoroughly understood the sources and types of

related data and we have gathered, collected some useful data. We are focusing on

finding the right insights from data.

The data we receive from twitter streaming is in Json format. This data needs to be

understood before using it, selecting the fields form this data is one of the important
 19
aspects. We use twitter4j library which is embedded into TwitterUtils library for getting

specific fields out of the data. The raw Json looks as follows:
20
 21

4.3 Pre- Processing Data

The collected data were noisy, missing useful info and inconsistent [14]. We have almost

finished the Data preparation process. In this process, we have to check if there are empty

values or inconsistency in the data. The data should be in a consistent state to be

analyzed. To improve the efficiency of our analysis the data should be in a simple format.

Data mining is done on this data so to get efficient results the data must be processed by

removing redundancies. The data is made meaningful by deriving information from it

like deriving dates out of months, days and year. For example, Deriving the age of the

tweet by its date.

A schema of organized data can be seen as follows

Figure 9: DataFrame Schema

 22

4.4 Algorithm and Sentiment Analysis

Sentiment analysis is performed using Stanford’s Natural Language Processing library

[15]. It takes text input and sends this library to get the sentiment in return. This library

constructs a tree like structure out of the plain text passed to it, this structured is created

after cleaning the data and removing all the stop words.

Calling function SentimentAnalysisUtils with parameters as plain text from the tweet.

The returned value is saved into map so that the status id tagged with its sentiment and

can be used further. This is the code for filtering the

After we filter the code and get the plain text we pass it to the function which creates a

tree out of it for getting the sentiment score. The function is defined as follows:
 23

The score is counted by taking calculating the average i.e. by dividing the sum of score

by the size of sentence as each sub tree and each word has a score associated to it.
 24
The score is then used to get the weighted sentiment out of the status and then mapped to

the sentiment based on the score. Here we have stated that 0 score is very negative, 1 is

associated to negative, 2 is neutral, 3 is positive and 4 is very positive.

4.4 Developed Model

The model should be reviewed and checked before using it for obtaining results. The

results obtained after mining the data should be analyzed meticulously and interpreted by

experts in order to perform efficient data analysis.

Figure 10: Basic Workflow Model [16]

 25
Above figure describes the steps involved in our project. Basic flow diagram of how we

use our raw data to extract relevant information. The extracted information is saved into

case classes and then saved into data frames. The data frame is created using the structure

we need.

Case Class:

Constructing DataFrames from schema string. The data types of each column needs to be

specified while creating the schema of for data frame.

 26

A dataframe looks like a table. When we use show () method on a dataframe it looks like.

Figure 11: DataFrame

4.5 Saving to Database

I am using MySQL to store the huge data. MySQL is widely used and is very efficient in

storing tabular data. It is easy to store data into MySQL using JDBC connection. The

connection needs a driver which then makes a connection to the database and pushes the

data into desired table.

 27

Every time a new data frame is created it gets pushed and appended to the existing table

in the database. The driver needs MySQL server URL and password along with table and

database name to store data. In order for the data to be isolated we need to set the

isolation property to read committed.

 28
Chapter 5

EXECUTION

5.1 Building Project

Launching the project on spark cluster needs a jar file. I use SBT as the package builder

for generating a .JAR of the project. SBT is widely used for building Scala and Java

projects. This JAR downloads all the dependencies needed for execution. This jar is

called FAT JAR or UBER JAR which contains all the code segments and dependencies.
 29

5.2 Submitting Job

Spark cluster needs to be started before submitting any job to the cluster. All worker

nodes to be initiated while starting the cluster. Spark contains shell code for starting

master, starting workers, stopping master, stopping workers, starting all, stopping all, etc.
30
 31
5.3 Tableau

Setting up tableau with the data source [12]. Tableau has two modes which are live and

extract.

Figure 12: Tableau Connection

 32
Chapter 6

USE-CASES AND RESULTS

6.1 IPhone vs Samsung Note7

One of the most observed dilemmas for common people is comparison between Apple

IPhone and Samsung (android phones). [17] The application would filter the live tweets

based on the filters set. Getting all the features from the clean data sentiment analysis

would be done on the data set. This comparison is made based on sentiments base on the

tweet, their locations and other factors. This comparison gives us a brief idea of how,

what and where people think on these products (Apple and Samsung)

This would answer questions like:

1. Where is IPhone most Popular?

2. Where is android most Popular?

3. What do peoples sentiments about IPhone and android?

4. How many people on an average from 50,000(Tweets collected) use which

phone?

5. How many influential people follow either of them?

6. Are there any trend changes after the recent changes in models?

The observations below are based upon data collected on different dates and the total size

of data base is 15000 rows plus.

 33

Figure 13:Density, Sentiment and Location Graph for Iphone vs Samsung (New York,

USA)
 34

Figure 14: Density, Sentiment and Location Graph for Iphone vs Samsung (Washington,

USA)
 35

Figure 15: Popularity Graph Iphone v/s Samsung Note7

 36

Figure 16: Location Based Sentimental Analysis

 37
6.2 PlayStation vs Xbox vs Nintendo

Other confusing task found amongst young generations is game and console selections.

There are many game stations available in today’s world but the Sony PlayStation,

Windows Xbox and Nintendo are the market leaders. All of them have their own follower

base and come with their own specialties. To solve this dilemma, I picked up this use

case for getting heat map based on location and sentiment.

The total size of dataset collected for this use case is 200000 plus.
 38

Figure 17: Heat Map for Gaming Consoles (Canada)

 39

Figure 18: Heat Map for Gaming Consoles (India)

 40
Chapter 7

FUTURE WORK

The world is currently moving towards mobile computing. Mobile applications have

become an important factor for a products success. It’s also very convenient to access

mobile applications on the Go. In future a mobile application could be developed which

would give instant reviews as soon as you set the filers as you desire. The task for mobile

computing would be optimization of the cluster on such a small device with minimal

hardware support and visualization.

 41
Chapter 8

CONCLUSION

I have learnt a lot of technologies while developing this project. The problems I faced

made me think deeper and out of the box. The vibrant and multidimensional visualization

would help consumers select a better product and avail the benefits of real time review

system based on location, public view and trend. It will also benefit businesses to get

feedback based on it there would be a huge scope for development of their product or

service.

During the course of implementation, I have gained in-depth knowledge in field of big

data analytics. I got to learn concepts like cluster computing on Apache spark which is a

very powerful platform and a trending platform for cluster computing, language like

Scala which is object oriented as well as functional language provides a lot more

flexibility for writing optimized code in smaller number of lines and has great capability

with the spark cluster. Tools like SBT and Tableau which are used wide across the

technological companies.

At last but not the least is I have learned how to overcome problems and develop a

product that would help people in their lives. The experience has made me technically

skilled and developed my thinking power around the problems.

 42
BIBLIOGRAPHY

[1] Forbes. 12 Big Data Definitions. [Online]. Available:

http://www.forbes.com/sites/gilpress/2014/09/03/12-big-data-definitions-whats-

yours/#6490f23421a9. Accessed in September 2014.

[2] Wikipedia. Big Data. [Online]. Available:

https://en.wikipedia.org/wiki/Big_data. Accessed in November 2013.

[3] Research Gate. The five V’s of Big Data. [Online]. Available:

https://www.researchgate.net/figure/281404634_fig1_Figure-1-The-five-V's-of-

Big-Data-Adapted-from-IBM-big-data-platform-Bringing-big. Accessed in

September 2015.

[4] Slide Share. Big Data Characteristics. [Online]. Available:

http://www.slideshare.net/venturehire/what-is-big-data-and-its-characteristics.

Accessed in July 2013.

[5] Wikipedia. Twitter. [Online]. Available: https://en.wikipedia.org/wiki/Twitter.

Accessed in November 2012.

[6] Spark. Spark Overview. [Online]. Available: http://spark.apache.org/docs/2.0.1/.

Accessed in October 2013.

[7] Spark. Cluster Mode Overview. [Online]. Available:

http://spark.apache.org/docs/latest/cluster-overview.html. Accessed in October

2013.
 43
[8] Spark. Spark Streaming Programming Guide. [Online]. Available:

http://spark.apache.org/docs/2.0.1/streaming-programming-guide.html. Accessed

in October 2013.

[9] Spark. Spark SQL, DataFrames and Datasets Guide. [Online]. Available:

http://spark.apache.org/docs/2.0.1/sql-programming-guide.html. Accessed in

October 2013.

[10] Spark. Spark SQL, DataFrames and Datasets Guide. [Online]. Available:

http://spark.apache.org/docs/2.0.1/sql-programming-guide.html. Accessed in

October 2013.

[11] Wikipedia. SBT. [Online]. Available:

https://en.wikipedia.org/wiki/SBT_(software). Accessed in September 2013.

[12] Wikipedia. Tableau. [Online]. Available:

https://en.wikipedia.org/wiki/Tableau_Software. Accessed in April 2013.

[13] Spring by Pivotal. Registering an application with Twitter. [Online]. Available:

https://spring.io/guides/gs/register-twitter-app/. Accessed in May 2014.

[14] Big Sky. The Data Analysis Process. [Online]. Available:

http://www.bigskyassociates.com/blog/bid/372186/The-Data-Analysis-Process-

5-Steps-To-Better-Decision-Making. Accessed in March 2013.

[15] Stanford CoreNLP. A Suite of core NLP Tools. [Online]. Available:

http://stanfordnlp.github.io/CoreNLP/. Accessed in August 2014.

[16] Recommender Systems. Data Mining. [Online]. Available: http://recommender-

systems.readthedocs.io/en/latest/datamining.html. Accessed in March 2013.