�ݺ�ߣ

Opinion Mining
Mohammed Al-Mashraee
Corporate Semantic Web (AG-CSW)
Institute for Computer Science,
Freie Universität Berlin
almashraee@inf.fu-berlin.de
http://www.inf.fu-berlin.de/groups/ag-csw/

AG Corporate Semantic Web
Freie Universität Berlin

Agenda
 Introduction
 Facts and Opinions and motivations
 Saentiment Analysis (SA) or Opinion Mining
 Why Sentiment Analysis
 What is Sentiment and Sentiment Analysis
 Sentiment Analysis Applications
 Sentiment Analysis Components
 Sentiment Analysis Model
 Sentiment Analysis Levels
 Document Level
 Sentence Level
 Feature Level
 Sentiment Analysis Approaches
 Supervised Approach
 Unsupervised Approach
 Case Studies

2

Agenda
 Introduction
 Saentiment Analysis (SA) or Opinion Mining
 Document Level
 Sentence Level
 Feature Level
 Case Studies

3

Types of data
 Facts/Objective
 Expressess facts
 E.g.,

− I bought a new car yesterday.
− This is a Canon Camara.

 Opinions/Subjective
 Expressess personal feelings or beliefs.
 E.g.,
− This Camara ist amazing.
− The resolution of this camera is fantastic.


5

Everyone needs it

Firms

Education

Health Care

Politics

Individuals

7

Making Decisions
I need
to buy a
camera

I need to
attend a
movie

I need to
Know about
this medicine

Why do
you vote
for X?


Opinion Sources:
− Parents
− Friends
− Neighbors

8

Making Decisions
How satisfy
our customers
are?

What about
our new
products?

How to face
competitors
and improve
products?

Opinion Sources:
− Surveys
− Focus Groups
− Opinion Polls


9

Search Engines


10

More interesting - Web 2.0
− social media Networks:
− Reviews:

− Blogs


11

Agenda
 Introduction

 Sentiment Analysis (SA) or Opinion Mining
 What is Sentiment and the Sentiment Analysis
 Document Level
 Feature Level

• Supervise Approach
• Unsupervised Approach

 Case Studies


12

Why Sentiment Analysis (SA)?

http://www.google.com/shopping

14

OM Synonyms
[Arti Buche, 2013]









Sentiment Analysis
Opinion Extraction
Sentiment Mining
Subjectivity Analysis
Affect Analysis,
Emotion Analysis,
Review Mining


15

What is Sentiment
Feeling, attitude, or opinions expressed by some
one towards something

16

Sentiment Analysis (SA)?
Sentiment analysis, also called opinion mining, is the field of study
that analyzes people’s opinions, sentiments, evaluations, appraisals,
attitudes, and emotions towards entities such as products, services,
organizations, individuals, issues, events, topics, and their attributes.
(Bing Liu 2012)

Text Mining

SA
Machine Learning
Machine Learning

Information Retrieval
Information Retrieval
Sentiment Analysis

Natural Language
Natural Language
Processing
Processing

Data Mining
Data Mining

Related areas of sentiment analysis

17

SA Applications
 Consumer Products and Services.
 Real-time Application Monitoring using
Twitter and/or Facebook.
 Financial Market Services.
 Political Elections.
 Social Events.
 Healthcare.
 Web advertising.


19

Opinion Mining Components
Opinion Holder (source)
The person or organization that
holds a specific opinion on a particular
object/target.
Opinion Target
A product, person, event,
organization, topic or even an
opinion.

Source

Opinion

Target

Opinion Components

Opinion Content
A view, attitude, or appraisal on an
object from an opinion holder.

21

Agenda
 Introduction

 Document Level
 Supervised Approaches
 Unsupervised Approaches
 Sentence Level
 Construct a Sentiment Lexicon
 Manually-based Method
 Dictionary-based Method
 Corpus-based Method
 Feature Level
 Feature Extration
 Feature Sentiment Orientation Detection


22

Opinion Mining Model:
[Bing Liu, ]

An object O is an entity which can be a product,
topic, person, event, or organization. It is associated
with a pair, O: (T, A), where T is a hierarchy or
taxonomy of components (or parts) and subcomponents of O, and A is a set of attributes of O.
Each component has its own set of sub-components
and attributes.


24

Opinion Mining Model
 The general term object is used to denote the entity that has
been commented on.
 An object has a set of components (or parts) and a set of
attributes.
 Each component may also have its sub-components and its
set of attributes, and so on.
Camera X

Lens

Baterry

Picture

Zoom

Camera X and ist related features

25

 An opinion is a quintuple (ej, ajk, soijkl, hi, tl)
such that






ej is the target entity,
ajk is an aspect of the entity ej ,
hi is the opinion holder,
Tl is the time when the opinion is expressed, and
soijkl is the sentiment orientation of opinion holder hi
on feature ajk of entity ej at time tl


26

 Explicit Attributes
Appears in the sentence as nouns or noun phrases.
E.g.,
The resolution of this camera is great.

 Implicit Attributes
Adjectives, adverbs, verbs, verb phrases, etc. that indicate
aspects
implicitly
E.g.,
This laptop is heavy.  (weight).
I installed the software easily.  (installation)


27

Agenda
 Introduction
 Document Level
 Sentence Level
 Feature Level
 Case Studies


28

Document level
Assumptions:
 Single

object for each document
 Single opinion holder

Task:
Determine the overall sentiment orientation in
a document/post/review (positive, negative, neutral)


30

Document level
E.g.,
“I bought a new X phone yesterday. The voice
quality is super and I really like it. However, it
is a little bit heavy. Plus, the key pad is too soft
and it doesn’t feel comfortable. I think the
image quality is good enough but I am not sure
about the battery life…”


31

SA Levels
Sentence level
Assumptions:
 Single opinion holder
The opinion is on a single object
Tasks:
Subjectivity Classification (subjective, objective)
Sentence polarity (positive, negative, neutral)
Eg.,
This is my car
My car is good


32

SA Levels
 Document and sentence level sentiment
analysis is too coarse for most
applications.
 Review assigned positive polarity for a
particular object does not mean people
are totally agree with that object


33

SA Levels
Feature level:
Goal: produce a feature-based opinion summary of
multiple reviews
Task 1: Identify and extract object features that
have been commented on by an
opinion holder (e.g. “picture”,“battery life”).
Task 2: Determine polarity of opinions on features
classes: positive, negative and neutral
Task 3: Group feature synonyms

34

Example Review
Document-based
“I bought a new X phone yesterday. The voice
quality is super and I really like it. The video is
clear. However, it is a little bit heavy. Plus, the
key pad is too soft and it doesn’t feel
comfortable. The zoom is great. I think the
image quality is good enough. I am not sure
about the battery life…”

35

Example Review
Sentence-based
The voice quality is super and I really like it (- po)
The video is clear (–po)
However, it is a little bit heavy (–ne)
Plus, the key pad is too soft and it doesn’t feel
comfortable (-ne)
The zoom is great (- po)
I think the image quality is good enough (- po)
I am not sure about the battery life


36

Example Review
Feature-based
voice quality
video
However, it is
key pad
zoom
image quality
battery life

super and I really like it
(- po)
clear
(–po)
heavy
(–ne)
too soft and doesn’t feel comfortable (-ne)
great
(- po)
good enough
(- po)
not sure
(–ne/ neutral)


37

http://www.tech-blog.net/review-htc-sensation-xe-teil-2/
http://www.euro.com.pl/lustrzanki/canon-eos-600d-18-55-mm-isii.bhtml#opinie
http://www.buydig.com/shop/product.aspx?
sku=CNDRT3I1855&ref=cnet&omid=113&CAWELAID=819186542&
http://reviews.cnet.com/digital-cameras/canon-eos-rebel-t3i/45056501_7-34499702.html


38

Agenda
 Introduction
 Document Level
 Sentence Level
 Feature Level
 Case Studies


39

Supervised Approach

 Supervise Approaches
 Availability of big amount of data
 Data representation
 Training data
 Testing data
 Unsupervised Approaches


41

Unsupervised Approaches
• Sentiment words and phrases are the main
indicators of sentiment classification
(e.g., adjectives, adverbs, etc.).
• Does not require big amount of data sets


42

The state of the art Cont.
( Turney. 2002)

 PMI-IR but this time to classify reviews into recommended and
not recommended in three steps:
1. Extract phrases containing adjectives or adverbs.
2. Estimate the semantic orientation of each extracted phrase
PMI(word1;word2) =
log2(p(word1&word2)/p(word1)p(word2))
SO(phrase) = PMI(phrase; "excellent") - PMI(phrase; "poor").
3. Classify the review based on the the average semantic
orientation of the phrases.
 If the average semantic orientation is possitive then the review is
classied as recommended and vice versa.

43

How to sentiment analysis
1. Pre-processing steps
• Collect a large body of reviews in text form
• Tokenization: break them down to a word by word level,
where each word is tagged with a “part of speech” token that
classifies it.
• The “part of speech” tagging can identify punctuation,
adjectives, verbs, nouns, pronouns.
• Stop words removal (the, of, at, in, …)
• Stemming: Relate words to their roots
(e.g., played, plays, playing  Play)


44

How to sentiment analysis
2. Sentiment classification
Apply a classifier to specify the the polarity of the given reviews
 Naive Bayes
 Decision Tree
 SVM


45

References
B. Pang, L. Lee, and S. Vaithyanathan, Thumbs up?: sentiment classication usingmachine learning techniques," in Proceedings of
the ACL-02 conference on Empirical methods in natural language processing - Volume 10, EMNLP '02, (Stroudsburg, PA, USA),
pp. 79{86, Association for Computational Linguistics, 2002.
K. Dave, S. Lawrence, and D. M. Pennock, Mining the peanut gallery: opinion
extraction and semantic classication of product reviews," in Proceedings of the
12th international conference on World Wide Web, WWW '03, (New York, NY,
USA), pp. 519{528, ACM, 2003.
Harb, M. Planti, G. Dray, M. Roche, Fran, o. Trousset, and P. Poncelet, "Web opinion mining: how to extract opinions from
blogs?," presented at the Proceedings of the 5th international conference on Soft computing as transdisciplinary science and
technology, Cergy-Pontoise, France, 2008.
http://de.slideshare.net/KavitaGanesan/opinion-mining-kavitahyunduk00
Case study
http://inboundmantra.com/sentiment-analysis-of-tripadvisor-reviews-hotel-leela-kempinski-case-study/

47

�ݺ�ߣ

Om 2

More Related Content

Om 2