CORE Metadata, citation and similar papers at core.ac.

uk
Provided by ethesis@nitr

Opinion Mining
of
Online Customer Reviews

Patlammagari Gowtamreddy

Department of Computer Science and Engineering

National Institute of Technology Rourkela
Rourkela – 769 008, India
 Opinion Mining
of
Online Customer Reviews

Dissertation submitted in

May 2014

to the department of

Computer Science and Engineering

of

National Institute of Technology Rourkela

in partial fulfillment of the requirements

for the degree of

Master of Technology

by

Patlammagari Gowtamreddy

(Roll 212CS1362)

under the supervision of

Korra Sathya Babu

Department of Computer Science and Engineering

National Institute of Technology Rourkela
Rourkela – 769 008, India
dedicated to my Parents and inspiring guide...
 Computer Science and Engineering
National Institute of Technology Rourkela
Rourkela-769 008, India. www.nitrkl.ac.in

Dr. Korra Sathya Babu

Professor

May , 2014

Certificate

This is to certify that the work in the thesis entitled Opining Mining of Online Customer

Reviews by Patlammagari Gowtamreddy, bearing roll number 212CS1362, is a record

of an original research work carried out by him under my supervision and guidance in

partial fulfillment of the requirements for the award of the degree of Master of Technology

in Computer Science and Engineering. Neither this thesis nor any part of it has been

submitted for any degree or academic award elsewhere.

Korra Sathya Babu

 Acknowledgement

This dissertation, though an individual work, has benefited in various ways from several peo-

ple. Whilst it would be simple to name them all, it would not be easy to thank them enough.

The enthusiastic guidance and support of Prof.Korra Sathya Babu inspired me to stretch

beyond my limits. His profound insight has guided my thinking to improve the final product.

My solemnest gratefulness to him.

I am also grateful to Dr.S.K.Rath for his ceaseless support throughout my research work.

My sincere thanks to Prof.S.K.Jena for his continuous encouragement and invaluable advice.

Many thanks to my comrades and fellow research colleagues. It gives me a sense of happiness

to be with you all. Special thanks to Jitendra Kumar Rout and Pramit Mazumdar whose

support gave a new breath to my research.

Finally, I am grateful to all my friends for continuous motivation and encouragement. Last

but not the least to my family having faith in me and always supporting me.

Patlammagari Gowtamreddy
 Declaration

I Patlammagari Gowtamreddy of Computer Science and Engineering with Roll no 212CS1362

hereby declare that the project submitted by me is solely of my work and is not copied from

any other source where ever may available and It has not been previously submitted for any

academic degree. I had verified my thesis report through Turnitin software for plagiarism.

All sources of quoted information have been acknowledged by means of appropriate refer-

ences.

If in future my work was found to be plagiarized from any other persons

work, then in that situation I alone will be responsible for it.

Date: June 2014

Patlammagari Gowtamreddy

NIT Rourkela
Abstract

Customer Opinions play a very crucial role in daily life. When we have to take a decision,

opinions of other individuals are also considered. Now-a-days many of web users post their

opinions for many products through blogs, review sites and social networking sites. Business

organizations and corporate organizations are always eager to find consumer or individual

views regarding their products, support and service. In e-commerce, online shopping and

online tourism, its very crucial to analyze the good amount of social data present on the

Web automatically therefore, its very important to create methods that automatically clas-

sify them. Opinion Mining sometimes called as Sentiment Classification is defined as mining

and analyzing of reviews, views, emotions and opinions automatically from text, big data

and speech by means of various methods. In this thesis we are going to see how Apriori

frequent item set mining algorithm can be used for mining reviews from online reviews those

are posted by customers. Our main theme is to create a system for analyzing opinions which

implies judgement of different consumer products.

Keywords: Opinion Mining, Sentiment Classification, Apriori Algorithm, SentiWordNet,

Min-max Normalization, Frequent Words, Online Reviews.

vii
Contents

Contents i

List of Figures iii

List of Tables iv

1 Introduction 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Opinion Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.3 Sentiment Analysis or Sentiment Classification [9] . . . . . . . . . . . . . . . . 2

1.4 Customer Produced Content [1, 2] . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4.1 Review sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4.2 Blogs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4.3 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.5 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.6 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.7 Thesis Organisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Introduction to Opinion Mining 6

2.1 Need for Opinion Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Opinion Mining Terminology [3, 4] . . . . . . . . . . . . . . . . . . . . . . . . 7

i
 2.3 Steps in Opinion mining approach . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3.1 Document-Level Opinion Mining [5] . . . . . . . . . . . . . . . . . . . . 10

2.3.2 Sentence-Level Opinion Mining [6] . . . . . . . . . . . . . . . . . . . . . 13

2.3.3 Phrase-Level Opinion Mining . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.4 Feature Level Opinion Mining . . . . . . . . . . . . . . . . . . . . . . . 16

3 Proposed Approach 18

3.1 Extract Nouns, Adjectives, Verbs and Adverbs . . . . . . . . . . . . . . . . . 19

3.2 Identify the frequent words by using Apriori Algorithm. . . . . . . . . . . . . 19

3.2.1 Apriori Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.3 Sentiment Analysis on the frequent words using SentiWordNet . . . . . . . . . 21

3.4 Min-max Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4 Implementation and Results 23

4.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4.2 Extracting Nouns, Adjectives, Verbs and Adverbs . . . . . . . . . . . . . . . . 23

4.3 Extracting frequent words using Apriori Algorithm . . . . . . . . . . . . . . . 24

4.4 Sentiment Analysis on the frequent words using SentiWordNet . . . . . . . . . 25

4.5 Min-max Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.6 Comparison between star rating and Opinion Mining process . . . . . . . . . . 28

5 Conclusion and Future Work 29

Bibliography 31
List of Figures

3.1 Proposed Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.1 dataset1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.2 dataset2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.3 Extracting Nouns, Adjectives, Verbs and Adverbs . . . . . . . . . . . . . . . . . 25

4.4 Extracting Nouns data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.5 Extracting frequent words using Apriori Algorithm . . . . . . . . . . . . . . . . . 26

4.6 Using SentiWordNet for finding polarity of a word . . . . . . . . . . . . . . . . . 26

4.7 Comparison between star rating and Opinion Mining process . . . . . . . . . . . 28

iii
List of Tables

4.1 Normalized Polarity values of Star ratings and Sentiwordnet scores . . . . . . . . 27

iv
Chapter 1

Introduction

1.1 Introduction

In this chapter, we give a brief introduction about what is Opinion Mining and how it can be

performed. Opinion Mining or Sentiment classification involves building a system to make

use of reviews posted by the users and opinions that are expressed in blogs and forums as

comments and reviews and sometimes may be as tweets about the product.

1.2 Opinion Mining

Opinion Mining is the science which combines techniques of computational linguistics and

information retrieval and is concerned with the opinions expressed rather than topics in the

text. Opinions are written on many things example a product, a topic, an individual, etc. In

opinion mining task we identify the orientation of opinion by the holder towards any object

which may be a collection of features or components or attributes.

Opinion mining [7, 8] might be valuable in a few ways. For instance, in adver-

tising, it tracks and judges the achievement rate of a commercial crusade or launch of new

item, focus prevalence of items and administrations with its forms additionally let us know

1
CHAPTER 1. INTRODUCTION 2

about demographics which like or hate specific characteristics. Case in point, a survey may

be around a computerized Polaroid may be comprehensively positive, yet be particularly

negative about how overwhelming it is. The seller gets overall picture of general opinion

than studies and centre gatherings, if this sort of data is indentified in a methodical manner.

1.3 Sentiment Analysis or Sentiment Classification [9]

What do other individuals think has dependably been a vital component in choice making

methodology. Sentiment Analysis or Sentiment Classification is the methodology to nat-

urally focus the sentiments communicated in a bit of plain content utilizing some regular

automated preparing systems. To be specific, term Sentiment is exceptionally wide and it

constitutes feelings, opinions, dispositions, particular encounters, and so forth. In this the-

ory, we speak just about the opinions communicated in writings which are composed in texts

which are written in human readable natural language, in social media.

Sentiment analysis is a procedure for following the views of the clients around a

specific item or subject. Sentiment analysis, which is likewise called opinion mining, in-

cludes in building a framework to gather and look at opinions about the item made in

blog entries, remarks, audits or tweets. Sentiment analysis might be helpful in a few ways.

Case in point, in showcasing it helps in judging the achievement of a notice crusade or new

item dispatch, figure out which forms of an item or administration are prominent and even

recognize which demographics like or aversion specific attributes.

CHAPTER 1. INTRODUCTION 3

1.4 Customer Produced Content [1, 2]

The web has all of a sudden changed the way how individuals express their perspectives

and opinions. They can now post audits on different sites, take an interest in discourse

on different discussions, compose a site depicting their experience, redesign their status

on social sites like Linkedin, Facebook, Google+ and etc. This information on audit sites,

exchange gatherings, websites, and interpersonal organizations might be combined and called

as customer produced substance. Each of these customers created substance has their special

property. In this thesis, we focus on product reviews given by the customers in various blogs,

review sites and forums.

1.4.1 Review sites

There are millions of sites where customers can compose views, reviews and opinions about

products, movies, and so on. A number of online shopping sites like Amazon, Flipkart allow

their clients to compose reviews on the product. These reviews impact the choice of the new

client who needs to purchase the item. Sites like Epinions, is an accumulation of reviews and

opinions from different products like vehicles, books, hardware, software and so forth. There

are a numerous sites like Rotten Tomatoes, Imdb which simply have film reviews of wide

sort and dialect space. These sites are extremely helpful from the point of view of opinion

mining.

1.4.2 Blogs

A Blog is a page where an individual or gathering of customers record opinions, data, and

so on all the time. Web blogs are composed on numerous various subjects like governmental

issues, games, travel and even items. Be that as it may, the nature of the content produced

from these sources is by generally noisy and poor. There are numerous web entrances like
CHAPTER 1. INTRODUCTION 4

blogspot, wordpress, and so on where any customer can make a record and post a web blog.

Numerous item and news sites likewise have a committed websites segment where limited

approved customers or users can post online reviews.

1.4.3 Dataset

We can take Dataset from many websites like Amazon, myntra and other online shopping

sites for performing Opinion Mining process which includes analyzing the reviews that are

posted by the customers or users on various products.

1.5 Motivation

Interest for sentiment analysis and opinion mining is expanding numerous fold. Customers

use web as a verbal exchange to express their choices focused around the opinions communi-

cated by others. The social media associate the whole world and are one of the explanations

behind data over-burden on the web. Twitter has hundreds of millions users who handle very

nearly half a million tweets for every day, normal of thousands of tweets for every second.

These tweets are posted on different dialects not simply English. These bewildering volumes

of upgrades call for a mechanized analysis of this text. There are numerous structures in

which customer produced substance is distributed on Internet. This inspired us to devise

systems to handle each of these various types of information and concentrate some helpful

data.

1.6 Objectives

Given an object and a collection of reviews on it, our objectives are

• Extract Nouns, Adjectives, Verbs and Adverbs on the remaining reviews by using dic-

tionary approach.
CHAPTER 1. INTRODUCTION 5

• Identify frequent words by using Apriori frequent item set mining Algorithm.

• Perform Sentiment Analysis on the frequent words using SentiWordNet.

• Provide visualization.

1.7 Thesis Organisation

The rest of the thesis contains various chapters as mentioned below:

Chapter 2, we discuss about the terminology and different methods used in opinion mining.

Then we outline several previous research efforts on opinion mining that attempts to develop

a method of extracting opinions.

Chapter 3, we present our approach towards opinion mining from customer reviews.

Chapter 4, we give the results we got for the data set containing customer reviews which

are posted on online review sites and we made the comparison between the results we got

for Apriori frequent item set mining algorithm using SentiWordNet and for star ratings that

we got from customer reviews.

Chapter 5, finally we conclude this thesis with future work.

Chapter 2

Introduction to Opinion Mining

A large portion of the current content handling systems work with authentic data. The

web holds enormous volumes of opinionated content. Web users express particular emotions

and opinions on very nearly anything at review sites, blogs, and forums and so on. This

important data is freely accessible for internet clients. The substantial gathering of opinions

on the Web makes it extremely tough to get helpful data effectively. Perusing all reviews

and emotions to settle on an educated choice is a much time taking task. Perusing distinc-

tive and potentially even conflicting opinions composed by diverse commentators may make

organizations, users and customers more confused. These needs have made another line of

examination on mining user and customer opinions, which is called opinion mining.

Opinion mining is the part of study that dissects individual opinions, sentiments,

assessments, mentality, and feelings from written text. It has pulled in a number of analysts

from distinctive areas of exploration including NLP, information mining, machine learning,

phonetics, and even social science. In the current chapter, we first examine need for opinion

mining and afterward characterize the terminologies utilized within this study. In the coming

6
CHAPTER 2. INTRODUCTION TO OPINION MINING 7

sections, we talk about general opinion mining tasks and present a concise survey of the

current and related deals with each one.

2.1 Need for Opinion Mining

With the growth of online social networking sites, for example, forums, review sites, blogs,

and micro blogs, the enthusiasm towards opinion mining has expanded essentially. Today on-

line opinions have transformed into a sort of virtual profit for business organizations looking

to market their items, recognize new trends and deal with their position. Many organizations

are currently utilizing opinion mining systems to track customer inputs in online shopping

sites and review sites.

Opinion mining [10] is additionally helpful for organizations to analyze customer

opinions on their products and features. While product attributes are clearly mentioned,

discovering the primary cause behind low profit needs much focus on all the more on indi-

vidual customer views on such characteristics. Opinion mining is an amazing method for

taking care of numerous business trends identified with deals administration, status man-

agement, and advertising. Additionally, organizations may have the capacity to perform

pattern prediction in deal by following customer perspectives.

2.2 Opinion Mining Terminology [3, 4]

In this segment we define the various terms used in the opinion mining.

Fact: A fact is that which has truly happened or is really the case.
CHAPTER 2. INTRODUCTION TO OPINION MINING 8

Opinion: An opinion is a view or judgement formed about something, not necessarily

based on fact or knowledge.

Subjective Sentence: A sentence or a text is subjective or opinionated if it actually

indicate ones feelings.

Objective Sentence: An objective sentence indicates some facts and known information

about the world.

Item: an individual article or unit, especially one that is part of a list, collection, or set.

Review: A review is a text containing a sequence of words thats has opinions of cus-

tomer for a specific item. A review may be subjective or objective or both.

Known Aspects: Known aspects are default aspects provided by the certain website for

which users separately give ratings.

Sentiment: Sentiment is a polarity term that implies to the direction in which a con-

cept or opinion is expressed. We use sentiment in a more specific sense as an opinion about

an aspect. For example, excellent is a sentiment for the attribute ‘battery life’ in the sentence

“This mobile has excellent battery life”.

Opinion Phrase: An opinion phrase is a pair of head term and modifier. Usually the

head term is a candidate aspect, and the modifier is a sentiment that expresses some opinion
CHAPTER 2. INTRODUCTION TO OPINION MINING 9

towards this aspect.

Opinion Polarity: Opinion Polarity or Subjectivity Orientation denotes the polarity ex-

pressed by the user or customer in terms of numerical values.

• Polarity: Polarity is a two way orientation scale. In this, a sentiment can be either

positive or negative.

• Rating: Most of the reviewing websites use star ratings for expressing polarity, pre-

sented by stars in the range from 1 to 5 which are called ratings.

Overall Rating: All the online shopping websites ask customers to give an overall rating

for the product that they already bought mentioning the overall quality of the used item.

Part-of-Speech (POS) Tag: POS tagging is very useful in Opinion Mining process. When

we need to analyse a document or a sentence first we have to extract the subjective infor-

mation from the document or that particular sentence. POS tagging helps us in getting

subjective words like Nouns, Verbs, Adverbs and Adjectives. After extracting these words

we can perform various actions on these and we can come to a conclusion.

2.3 Steps in Opinion mining approach

In this segment we show a survey of the current and related tasks in brief at opinion mining

proposed in the existing methods. For a thorough survey, we sort opinion mining proce-

dure into three general classes, yet before we talk about our arrangement, we introduce

current order constructions experienced in the writing. Pang et al. [11] group the significant

issues of opinion mining into three classes: sentiment polarity identification, subjectivity
CHAPTER 2. INTRODUCTION TO OPINION MINING 10

detection, and joint topic-sentiment analysis. Some experts additionally characterizes three

mining methods for opinionated content in his book. He further grows this classification in

his handbook as: sentiment and subjectivity characterization, aspect based opinion mining,

sentiment analysis of similar sentences, opinion search and data retrieval, and opinion spam

identification. At last, in his latest book he characterizes three general classifications for

opinion mining tasks: document level, sentence level, and phrase level.

There is one more interesting study in the opinion mining tasks classification called

feature level. We discuss about feature level opinion mining in brief in this section.

• Document level opinion mining

• Sentence level opinion mining

• Phrase Level opinion mining

• Feature Level opinion mining

2.3.1 Document-Level Opinion Mining [5]

Document level tasks primarily concerns with classification issues where the available docu-

ment has to be arranged into a set of predefined classes. In subjectivity analysis a document

is divided as subjective or objective. In sentiment analysis, a document can be classified as

positive or negative or neutral depending upon the polarity of subjective information that
CHAPTER 2. INTRODUCTION TO OPINION MINING 11

is present in the document. Opinion quality and support evaluation makes decision whether

an opinion is useful or not and opinion spam identification groups and divides opinions as

spam and not spam.

Subjectivity Analysis

Subjectivity Analysis [12, 13] refers to finding whether the given document makes an opinion

or not. To be precise, whether a document or text is objective or subjective. We consider

this problem generally as a classification problem. Many of the current methods uses super-

vised learning, even though we have few unsupervised methods. One of the works in this

area given by Wiebe et al. [14] does subjectivity analysis using the naive Bayesian classi-

fier. Succeeding research uses other learning algorithms for finding subjective text. Future

research has been mainly focused on developing automated process for subjectivity analysis.

One of the tough tasks in subjectivity classification is the human effort involved in labelling

training examples as subjective or objective.

Sentiment Analysis

Sentiment classification in default takes the given review is subjective and expects to discover

the general sentiment of user in the content. Suppose, we have a product review, it figures

out if the review is of positive polarity or negative polarity. Sentiment classification, as other

way to subjectivity examination, does not typically require manual exertion for training the

data. Preparing information utilized as a part of sentiment classification are generally online

product reviews that already been marked by reviewers. Current works preferably focuses
CHAPTER 2. INTRODUCTION TO OPINION MINING 12

on supervised learning approach to sentiment classification. We have three existing ma-

chine learning methods namely naive Bayes, Maximum Entropy classification, and Support

Vector Machine (SVM) to classify movie reviews as positive polar or negative polar. We

infer that the standard machine learning methods perform far better than human produced

approaches. We have few unsupervised methods for dividing reviews.

Opinion Quality and support evaluation [15]

The issue of consequently assessing the support of online reviews has additionally pulled

in expanding consideration. Most past works endeavour to foresee the supportiveness of

reviews by utilizing a set of selected features, like literary and social features and taking in

a capacity of these features for foreseeing survey accommodation. Literary features incorpo-

rate features that are focused around content facts, for example, length of the survey, the

normal length of a sentence, rate of things or descriptive words, and so forth. Social features,

then again, are identified with the making of the survey and are concentrated from his social

setting, for example, the amount of past reviews by the creator, in-degree and out-level of

the creator in the informal community, past normal rating for the creator, and so forth.

Opinion Spam identification

Web spam is very well known to most of the individuals. In the view of opinion, we have a

comparative spam issue. The development of customer created substance inspires more in-

dividuals to discover and read opinions on the Web. Positive opinions can bring about profit

and money related addition or acclaim for the maker or merchants, and negative opinions,
CHAPTER 2. INTRODUCTION TO OPINION MINING 13

then again, can have the opposite effect on them. This effect additionally gives great moti-

vating forces for composing spam reviews called as opinion spam. Spam opinions attempt

to deliberately misdirect customers by giving undeserving positive or negative opinions to

some target things so as to push the thing as well as by giving undeserving negative opinions

to some different things to harm their position.

Spam identification could be planned as a characterization issue with two classes:

spam and no spam. The principle assignment is to discover the situated of viable information

features for building the model. Jindal et al. [16, 17] distinguish some text based and social

features for taking in a relapse model to gauge the likelihood of each one review being a

spam. In distinctive systems for distinguishing spam reviews utilizing conduct of reviewers

are proposed. There are likewise a few deals with recognizing opinion leader, and finding

gathering spammers for reviews.

2.3.2 Sentence-Level Opinion Mining [6]

The problem at this measure everything refers to sentences. In opinion data extraction

and recovery and opinion question answering, sentences are generally placed and positioned

focused around some criteria. Opinion rundown intends to select a set of sentences which

outline the opinion all the more exactly. At long last, opinion mining in relative sentences

incorporates recognizing similar sentences and concentrating data from them.

Opinion Search and Retrieval

Conventional Web quest is exceptionally essential for online customers. Opinion retrieval

[18] will be likewise of incredible utilization. Looking the customer produced substance on
CHAPTER 2. INTRODUCTION TO OPINION MINING 14

the Web empowers review readers to discover opinions on any topics. Opinion data extrac-

tion methods are for the most part issued to discover popular opinion on a specific thing

or a part of the thing. For instance, to discover general opinion on a multimedia mobile, a

customer may raise the problem “mobile battery life”.

With respect to the classification task, web search tools generally rank WebPages

focused around correctness and certain score. The presumption is that the top positioned

pages hold sufficient data to fulfil the customers data need. In any case, this suspicion is not

genuine in the concept of opinions. The top positioned documents just speak to the opinions

of few persons not people in general. Current positioning techniques use distinctive crite-

ria to reflect general customer opinion. The system proposed in the behavioural model of

shoppers utilizing budgetary methodology for positioning items. Likewise in different tasks,

review quality, content detail, customer inputs are granted as metrics of positioning.

Opinion visualization [19]

The objective of opinion visualization is selecting entire sentences to make a short pas-

sage outline. Provided for a few reviews, a review outline system yields content comprising

requested sentences. Existing methodologies concentrate on selecting sentences so that the

outline incorporates vital data of the reviews.

Opinion Question Answering

Question Answering is one of the useful step in opinion mining. Opinion question answering
CHAPTER 2. INTRODUCTION TO OPINION MINING 15

process [20] gives a way to answer the decision based queries using the customers opinions

about the specific product. There is lot of difference between traditional question answering

and opinion question answering process. Opinion question answering is rather complicated

when compared to traditional question answering process. For instance consider the ques-

tion which is the best android mobile available in the market today? and some questions do

target on semantic opinion like You suggest me to buy this camera or not? Opinion question

answering is generally much time taking process and can be considered as virtual. Business

Organizations compare and classify the polarity of reviews and opinion question and answers.

The methods that are already in process creates a method that classifies between subjective

and objective at document level and then we get polarity of those by using various sentiment

classification approaches.

Opinion mining in similar sentences

Opinion mining in similar sentences in similar sentences refers to comparing two products

of similar category or sometimes we may compare only certain attributes of the different

products. Opinion mining in comparative sentences [21] demands the use of words like best,

better, larger, smaller and so on. These approaches are considered as classification problems

and generally require some machine learning techniques and training a classifier.

2.3.3 Phrase-Level Opinion Mining

Phrase level mining [5] came into picture because document level mining and sentence level

mining approaches cant find accurately what actually users likes and they does not like.

Phrase level opinion mining looks for sentiments on features of products.

CHAPTER 2. INTRODUCTION TO OPINION MINING 16

Aspect-Based Opinion Mining

In the recent days, lot of researchers are showing a great interest in aspect based opin-

ion mining [22]. We have a few approaches to retrieve aspects from comment and reviews.

Many of these methods use full text comments and reviews, where as a few approaches have

taken advantage of comments and reviews those are in sequential pattern.

2.3.4 Feature Level Opinion Mining

Feature level opinion mining comes into picture when a customer or user looking for feedback

of certain feature or attribute of a product rather than total feedback of the product. We

see many customers interested in only certain features of some products rather than the

whole product like some people look for a mobile that has excellent battery life and they are

not concerned with other features like camera clarity, music clarity and so on. In situations

like mentioned in this section feature level opinion mining helps a lot for extracting polarity

information for a particular feature or attribute from a product.

CHAPTER 2. INTRODUCTION TO OPINION MINING 17

Extract object features

In this step, we extract product features from comments and reviews. We may follow any

approach for doing this, but frequently used method is to extract nouns and noun phrases

there by getting actual features of the product.

Determine polarity of opinions on features

After extracting product features from reviews and comments using nouns and noun phrases,

we perform semantic analysis on the resulting features which gives information about the

each and every feature which customer liked and which he did not like.

Group feature synonyms

In some scenarios, we may end up a situation like we have various synonyms for same

attribute and feature from a product. By grouping those feature synonyms, our task be-

comes easier for performing opinion mining process. Suppose if a customer has written a

review mentioning that “this mobile can stand by for 2 days without battery recharging”,

this statement actually refers to hidden feature called ‘battery life.

Chapter 3

Proposed Approach

In this section, we discuss about proposed approach for performing opinion mining and what

techniques and Algorithms we are going to use to perform Opinion Mining and Sentiment

Analysis for getting useful information from online customer reviews. Our methodology in-

cludes the following steps:

• Extract Nouns, Adjectives, Verbs and Adverbs from the customer reviews by using dic-

tionary approach [23].

• Identify frequent words by using Apriori frequent item set mining algorithm [24, 25, 4].

• Perform Sentiment Analysis on the frequent words using SentiWordNet [26, 27, 28].

• Use Min-max normalization [29, 30] for normalizing the polarity of values that we got

using SentiWordNet and for normalizing the values of star ratings that we have taken

18
CHAPTER 3. PROPOSED APPROACH 19

Figure 3.1: Proposed Approach

from the customer reviews.

• Provide visualization.

The following figure shows an overview of proposed approach 3.1.

3.1 Extract Nouns, Adjectives, Verbs and Adverbs

To perform this step, we are using dictionary approach. After this step we get all the Nouns,

Adjectives, Verbs and Adverbs that are present in the customer reviews file. We use these

words for getting the frequent words in the next step.

3.2 Identify the frequent words by using Apriori Algorithm.

In this step, we get frequent words by using the various Apriori frequent itemset mining

algorithm.
CHAPTER 3. PROPOSED APPROACH 20

3.2.1 Apriori Algorithm

The following are the steps involved in Apriori Algorithm.

Assume for Ck and Lk

Ck denotes candidate itemset of k size

Lk denotes frequent itemset of k size

Important steps of algorithm are:

1) Initially get frequent set Lk−1

2) Join step: get Ck by doing cartesian product of Lk−1 with itself

3) Those itemsets which are of size (k-1) and those are not frequent should not be a subset

of a frequent itemset of size k, so those should be removed

4) Finally frequent set Lk has been achieved

CHAPTER 3. PROPOSED APPROACH 21

3.3 Sentiment Analysis on the frequent words using SentiWordNet

In this step we perform Sentiment Analysis on the frequent words that we got from Apriori

Algorithm by using SentiWordNet. It provides a value for each and every word.

Sentiment Analysis deals with the usage of automated techniques for anticipating the in-

troduction of subjective substance on text reviews or comments, with usage in various fields

that includes recommendation system and advertising, user intelligence and opinion retrieval.

Sentiwordnet is an opinion vocabulary and can be considered as extended from the Wordnet

database where each one term is connected with numerical scores demonstrating positive

and negative sentiment data. This examination shows the consequences of applying the

Sentiwordnet lexical asset to the issue of automated sentiment arrangement of customer film

reviews or comments.

3.4 Min-max Normalization

Min-Max normalization is the technique of taking data calculated in its own units and con-

verting it to a value between 0 and 1

We use normalization because star ratings values lies between 1 to 5 and word polarity

of SentiWordNet values lies between -1 and +1

Suppose we have some n rows with five variables, A, B, C, D and E, in the data. We

use variable B as an example for understanding the normalization concept in the calcula-

tions below. All the other variables in the rows are normalized in the similar way.
CHAPTER 3. PROPOSED APPROACH 22

The normalized value of Bi for variable B in the ith row is calculated as:

Bi − Bmin
normalised(Bi ) = (3.1)
Bmax − Bmin

where

Bmin = the least value for variable E

Bmax = the highest value for variable E

If Bmax = Bmin then Normalized (Bi ) is set to 0.5.

Chapter 4

Implementation and Results

This section contains results of our proposed Opinion Mining approach.

4.1 Dataset

The following picture shows the set of customer reviews for Bellagio hotel that is situated

in Las Vegas, USA. In the given file each row denotes one review with first column being

customer id, second column refers to star rating given by the customer for the product and

third column denotes the review written by the customer for the product.

The following 4.2 shows the set of Nouns, Adjectives, Verbs and Adverbs in the text file.

We search for these words in the data set and extract them from the text file containing text

file.

4.2 Extracting Nouns, Adjectives, Verbs and Adverbs

In the following figure 4.3 we see the extracted words using POS tagging from text file

containing customer reviews. The following figure represents a bar chart between number of

reviews and number of words extracted for different number of reviews.

23
CHAPTER 4. IMPLEMENTATION AND RESULTS 24

Figure 4.1: dataset1

Figure 4.2: dataset2

4.3 Extracting frequent words using Apriori Algorithm

This figure 4.5 shows the frequent words we got using Apriori Algorithm from the set of

words.
CHAPTER 4. IMPLEMENTATION AND RESULTS 25

Figure 4.3: Extracting Nouns, Adjectives, Verbs and Adverbs

Figure 4.4: Extracting Nouns data

4.4 Sentiment Analysis on the frequent words using SentiWordNet

The following figure 4.6 show how we can use SentiWordNet for finding the polarity of a

word.
CHAPTER 4. IMPLEMENTATION AND RESULTS 26

Figure 4.5: Extracting frequent words using Apriori Algorithm

(a) Sentiment analysis-1

(b) Sentiment analysis-2

Figure 4.6: Using SentiWordNet for finding polarity of a word

4.5 Min-max Normalization

We have star ratings value ranges between 1 to 5 and word polarity of SentiWordNet values

ranges between -1 and +1. So for mapping both the values on a scale of 0 to 1, we use

Min-max normalization.
CHAPTER 4. IMPLEMENTATION AND RESULTS 27

The following table shows the normalized polarity values for star ratings taken from

customer reviews and for SentiWordNet for different number of reviews 4.1.

No.of Reviews Star rating Sentiwordnet

10 0.6 0.48
20 0.57 0.5
50 0.52 0.47
100 0.42 0.39
200 0.38 0.36

Table 4.1: Normalized Polarity values of Star ratings and Sentiwordnet scores
CHAPTER 4. IMPLEMENTATION AND RESULTS 28

4.6 Comparison between star rating and Opinion Mining process

The following graph shows how the normalized polarity values gets varied for star ratings

values that are taken from customer reviews and for Opinion Mining process that uses Apri-

ori algorithm and SentiWordNet for different number of reviews.

Figure 4.7: Comparison between star rating and Opinion Mining process
Chapter 5

Conclusion and Future Work

Opinion mining has become a fascinating research area due to the availability of a huge

volume of user-generated content in review sites, forums and blogs. Opinion mining has

applications in a variety of fields ranging from market research to decision making to ad-

vertising. With the help of opinion mining, companies can estimate the extent of product

acceptance and can devise strategies to improve their product. Individuals can also use

opinion mining tools to make decisions on their buying by comparing competitive products

not just based on specifications but also based on user experience and public opinions. In

this thesis we have shown how Apriori frequent item set mining algorithm can be used in

Opinion Mining process. We have seen how normalized polarity varies for the values taken

from customer reviews star ratings and from values of opinion mining process that uses

Apriori algorithm and SentiWordNet.

Here, we see a few but important challenges for text analytics tasks like

opinion mining. It is very difficult to distinguish between objective and subjective informa-

tion, generally opinion words also occur in objective sentences, so it is very tough to handle

these challenges. Many times we see customers posting the reviews in the blogs or forums

29
CHAPTER 5. CONCLUSION AND FUTURE WORK 30

with a lot of spelling mistakes which our dictionary cannot catch them and resulting in less

accuracy of desired output. Most times of the times we see many spam blogs and spam

reviews posted by the users. If we consider these reviews for performing Opinion Mining,

we may get deviated from our desired results. So, lot of work has to be done in this field for

identifying spam blogs, considering spelling mistakes and for other challenges.
Bibliography

[1] Luı́s Sarmento, Paula Carvalho, Mário J Silva, and Eugénio de Oliveira. Automatic creation of a reference corpus

for political opinion mining in user-generated content. In Proceedings of the 1st international CIKM workshop

on Topic-sentiment analysis for mass opinion, pages 29–36. ACM, 2009.

[2] John Krumm, Nigel Davies, and Chandra Narayanaswami. User-generated content. IEEE Pervasive Computing,

7(4):10–11, 2008.

[3] Bo Pang and Lillian Lee. Opinion mining and sentiment analysis. Foundations and trends in information

retrieval, 2(1-2):1–135, 2008.

[4] Minqing Hu and Bing Liu. Mining and summarizing customer reviews. In Proceedings of the tenth ACM SIGKDD

international conference on Knowledge discovery and data mining, pages 168–177. ACM, 2004.

[5] Theresa Wilson, Janyce Wiebe, and Paul Hoffmann. Recognizing contextual polarity in phrase-level sentiment

analysis. In Proceedings of the conference on human language technology and empirical methods in natural

language processing, pages 347–354. Association for Computational Linguistics, 2005.

[6] Alexander Pak and Patrick Paroubek. Twitter as a corpus for sentiment analysis and opinion mining. In LREC,

2010.

[7] Andrea Esuli. Automatic generation of lexical resources for opinion mining: models, algorithms and applications.

In ACM SIGIR Forum, volume 42, pages 105–106. ACM, 2008.

[8] Richa Sharma, Shweta Nigam, and Rekha Jain. Supervised opinion mining techniques: A survey.

[9] Akshat Bakliwal. Fine-grained opinion mining from different genre of social media content. 2013.

[10] Samaneh Moghaddam and Martin Ester. Aspect-based opinion mining from online reviews. In Tutorial at SIGIR

Conference, 2012.

31
BIBLIOGRAPHY 32

[11] Bo Pang and Lillian Lee. A sentimental education: Sentiment analysis using subjectivity summarization based

on minimum cuts. In Proceedings of the 42nd annual meeting on Association for Computational Linguistics,

page 271. Association for Computational Linguistics, 2004.

[12] Ellen Riloff, Janyce Wiebe, and William Phillips. Exploiting subjectivity classification to improve information

extraction. In Proceedings of the National Conference On Artificial Intelligence, volume 20, page 1106. Menlo

Park, CA; Cambridge, MA; London; AAAI Press; MIT Press; 1999, 2005.

[13] Stephan Raaijmakers and Wessel Kraaij. A shallow approach to subjectivity classification. In ICWSM, 2008.

[14] Janyce M Wiebe, Rebecca F Bruce, and Thomas P O’Hara. Development and use of a gold-standard data set

for subjectivity classifications. In Proceedings of the 37th annual meeting of the Association for Computational

Linguistics on Computational Linguistics, pages 246–253. Association for Computational Linguistics, 1999.

[15] Anindya Ghose and Panagiotis G Ipeirotis. Estimating the helpfulness and economic impact of product reviews:

Mining text and reviewer characteristics. Knowledge and Data Engineering, IEEE Transactions on, 23(10):1498–

1512, 2011.

[16] Nitin Jindal and Bing Liu. Opinion spam and analysis. In Proceedings of the 2008 International Conference on

Web Search and Data Mining, pages 219–230. ACM, 2008.

[17] Nitin Jindal and Bing Liu. Review spam detection. In Proceedings of the 16th international conference on World

Wide Web, pages 1189–1190. ACM, 2007.

[18] Wei Zhang, Clement Yu, and Weiyi Meng. Opinion retrieval from blogs. In Proceedings of the sixteenth ACM

conference on Conference on information and knowledge management, pages 831–840. ACM, 2007.

[19] Lun-Wei Ku, Yu-Ting Liang, and Hsin-Hsi Chen. Opinion extraction, summarization and tracking in news and

blog corpora. In AAAI Spring Symposium: Computational Approaches to Analyzing Weblogs, volume 100107,

2006.

[20] Hoa Trang Dang and Karolina Owczarzak. Overview of the tac 2008 opinion question answering and summa-

rization tasks. In Proc. of the First Text Analysis Conference, 2008.

[21] Murthy Ganapathibhotla and Bing Liu. Mining opinions in comparative sentences. In Proceedings of the 22nd

International Conference on Computational Linguistics-Volume 1, pages 241–248. Association for Computational

Linguistics, 2008.

[22] Charu C Aggarwal and ChengXiang Zhai. Mining text data. Springer, 2012.
BIBLIOGRAPHY 33

[23] Xiaowen Ding, Bing Liu, and Philip S Yu. A holistic lexicon-based approach to opinion mining. In Proceedings

of the 2008 International Conference on Web Search and Data Mining, pages 231–240. ACM, 2008.

[24] Raffaele Perego, Salvatore Orlando, and P Palmerini. Enhancing the apriori algorithm for frequent set counting.

In Data Warehousing and Knowledge Discovery, pages 71–82. Springer, 2001.

[25] Christian Borgelt and Rudolf Kruse. Induction of association rules: Apriori implementation. In Compstat, pages

395–400. Springer, 2002.

[26] Bruno Ohana and Brendan Tierney. Sentiment classification of reviews using sentiwordnet. In 9th. IT & T

Conference, page 13, 2009.

[27] Andrea Esuli and Fabrizio Sebastiani. Sentiwordnet: A publicly available lexical resource for opinion mining. In

Proceedings of LREC, volume 6, pages 417–422, 2006.

[28] Stefano Baccianella, Andrea Esuli, and Fabrizio Sebastiani. Sentiwordnet 3.0: An enhanced lexical resource for

sentiment analysis and opinion mining. In LREC, volume 10, pages 2200–2204, 2010.

[29] Anil Jain, Karthik Nandakumar, and Arun Ross. Score normalization in multimodal biometric systems. Pattern

recognition, 38(12):2270–2285, 2005.

[30] Soo-Min Kim and Eduard Hovy. Determining the sentiment of opinions. In Proceedings of the 20th international

conference on Computational Linguistics, page 1367. Association for Computational Linguistics, 2004.