SENTIMENT ANALYSIS BASED APPROACHES FOR UNDERSTANDING USER - PowerPoint Presentation

Slide1

SENTIMENT ANALYSIS BASED APPROACHES FOR UNDERSTANDING USER CONTEXT IN WEB CONTENT

BY,

SOWMYA KAMATH,

ANUSHA BAGAL KOTHKAR,

KUMARI POORNIMA,

SHIVAM PANDEY

AND

ASHESH KHANDELWALSlide2

Introduction

Approaches to Sentiment Analysis

Sentiment Analysis ApplicationsCurrent development in Sentiment AnalysisConclusionFuture workReferences

overviewSlide3

Sentiment analysis, also known as opinion mining is the computational study of

opinions, sentiments

and emotions expressed in natural language for the purpose of decision makingSentiment analysis applies natural language processing techniques and computational linguistics to extract information about sentiments expressed by authors and readers about a particular subject, thus helping users in making sense of huge volume of unstructured

Web

dataFor example: In our day to day lives, we highly value the opinions of friends in making decisions about issues like which brand to buy or which movie to watch

introductionSlide4

Two types of textual information on webFacts

Opinions

Currently available search engines search for facts using machine readable informationIn today’s web, lot of opinioned text is available in various forms, for example, as reviews, blogs, news articles, discussion groups and social networking sitesAnalyzing opinions is very important for making decisions

Example-new cell phone

Sentiment analysis is currently a very significant trend in the area of natural language processing.introductionSlide5

Natural language processing involves giving artificial intelligence

to computers

and is concerned with promoting and understanding of human languages for machine’s useSentiment analysis extracts opinions, sentiments, and emotions from text and analyses them.Sentiment classification can be done at three levelsDocument level

S

entence levelFeature levelintroductionSlide6

A document can be classified

into two

classes positive and negative based on overall sentiment expressed by its writerClassification can be done based on four pairs of human emotions, namely,“Joy  Sadness”,

“Acceptance

 Disgust”“Anticipation  Surprise

”

and

“Fear



Anger”

Document level classificationSlide7

Sentence level sentiment analysis has two tasks, subjectivity classification and sentiment

classification

Information in a sentence can be of two types, objective information and subjective informationSubjectivity classification involves identifying whether the sentence is subjective or objectiveSentiment classification is further classifying the subjective information as positive or negativeFor example consider the following snippet

of text

- “I bought an iPhone a few days ago. It is a great phone.”Sentence level analysisSlide8

Feature level classification comprises of three main tasks

First step is to identify and extract the

featuresThe next step is to determine whether the opinions on the features are positive, negative or neutralFinal task is to group the feature synonymsIt has been found that document level and sentence level classification are not enough to identify each

and every

one detail about sentiments expressed in a document as sentiments may be expressed with respect to different features. For example, a phone may have a rating of 4 out of 5 for speed, 2 out of 5 ease of use, 3 out of

5 for

battery, etc.

Feature level extractionSlide9

Sentiment analysis classifies the opinions into positive and

negative

categoriesKnowing the reasons behind classifying the sentiment provides better perceptionThese reasons are called as sentiment topics associated with the sentimentThe proposed method collects web content and

extracts snippets

from them. Snippets are keywords like brand namesThen a sentiment score is calculated for each snippet based on which they are classified

into different categories to create a

sentiment taxonomy

Topics

related to each category

are then identified

Approaches to sentiment analysisSlide10

Hogenboom proposed a method

which considers

the negation scope and strength of a word while classifying whether a word has positive or negative effect on the sentence For example, let us consider two sentences “I am happy with your performance” and “I am not that happy with your performance”

The first sentence expresses a positive emotion If we just consider the negative keyword “not” then the second sentence would be equivalent to “I am not happy with your performance” which

is not

correct

If scope and strength of the

negative keywords

are considered while deciding its effect then it would give better resultsThe proposed approach uses two algorithms; the first one is used to calculate sentence

score for

each

word

In the second algorithm, the sentence

score is

calculated using the word sense and word score

with respect to each negative keyword. If the calculated sentence score is less than zero, then it is assigned to a negative class

Approaches to sentiment analysisSlide11

Methods to analyze sentiment include

machine learning

, statistical methods, building a knowledge base and identifying keywordsTo recognize effective information from text, sentence level analysis is required.Shaikh et al. developed a tool called SenseNet, that assigns numerical valence values and output sense value for each sentence.

The

input paragraph is divided into a set of sentences and each sentence is further divided into triplets. Valence values are assigned to the words in the triplet. These

triplets are then processed to calculate

the sentence

level sentiment

valence

Approaches to sentiment analysisSlide12

An overall view about

a document does not reveal the sentiments about

all aspects of a topic. For example, a person might be happy with the camera, music, games in his cell phone but its battery life may be a problem.Mapping the sentiment to the correct topic is quite a challenge The Sentiment Analyzer algorithm presented by Nasukawa

et.al. extracts the features related to a topic, and then extracts sentiments of each sentiment bearing phase. It associates this topic, feature and sentiment to the document

Classification based approachSlide13

An approach to classify news video stories and rank them has been presented by Chunxi

et.al.

In their approach, the stories were divided into two classes positive class and negative classThe algorithm forms two clusters - one containing positive adjectives and other containing negative ones. A graph based semi-supervised learning approach has been used for this purpose

Similarity

between words is calculated to find the sentiment words. The selected sentiment words are used as features for classificationFor the visual part, an

Affinity Propagation clustering approach is used

to determine

the ranking of the videos. A linking matrix

is used

to check similarity between videos. Both text

and visual information are combined to rank the video

Classification based approachSlide14

Cluster formed using cluster algorithmSlide15

A Support Vector Machine (SVM)

was used as the classifier

algorithm The other models used for comparison are Naïve Bayes classifier (NB), passive-aggressive classifier (PA), bigram (BI), word(WD), metadata (MT), affix similarity (AS), word emotion (WE) and Cui’s combined word n-grams (CN)The highest accuracy was achieved when the

models SVM

, BI, WD, MT, AS and WE were used togetherClassification based approachSlide16

Zhang et al used a method where, based on

keyword entered

by users, a sentiment graph of sentiment vectors of articles that keyword is plotted The sentiment graph gives an idea about inclination of articles towards various sentiments.Machine Learning in Document Level Classification is used to carry out sentiment analysis.

S

upervised methods can also be used support vector machine (SVM) - classifying reviews Naïve

Bayes

method – co-occurrence of each word

Maximum entropy classifier - weights

Entropy method

Classification based approachSlide17

Lacking conscious awareness of websites sentiment bias may result in

blind obedience

to the reported informationGiven a topic, Zhang et al proposed a system that extracts relevant subtopics and presents sentiment difference between different subtopics The system analyses a given sentiment in four dimensions, which is more similar to human emotion than

conventional positive-negative

sentiment and detects sentiment bias. In the system, articles are crawled and the part of speech tagging is done on themWeight for each extracted word from article is calculated

using

Sentiment analysis applicationsSlide18

where

N(w, Pi) is the number of times that word

w appears in article Pi, N(Pi) is the number of words extracted from Pi, N is the number of all collected news articles, and N(w) is the number of articles in which word w appears. a sentiment dictionary is constructed which contains a word and its sentiment value. Sentiment value consists of scale value and weight value for

four dimensions

.Sentiment analysis applicationsSlide19

Sentiment value is calculated using probability functions for each article. For a particular

year (Y

) edition for a particular newspaper, the number of articles which include any word in the set e of original sentiment words in Table 1 be df(Y, e), and the number of articles which include both target word w and any word in e be df(Y,e&w)Next interior division ratio and scale value is calculated using

Sentiment analysis applicationsSlide20

A word may appear in number of editions and number of times in various editions. To consider this, weight factor is calculated using

A sentiment value Oe(P) of article P on dimension e is calculated as follows

Sentiment analysis applicationsSlide21

Original sentiment words for the four dimensionsSlide22

Celikyilmaz et al.

considered that

twitter messages are of two types - polar and non polar (neutral).They present a probabilistic model based sentiment analysis approach for twitter messages. Their technique analyzes sentiments of polar text. As the twitter messages are human generated, it is very difficult to interpret its meaning correctly sometimes even by humans and there may be a lot of noise in it, in the form of slang, shorthand etc. The method proposed first does text

normalization followed

by pronunciation based clustering. For example, 4get is same as forget. Then, polarity lexicon extraction is done using a mixture model. The authors state that

this analysis

can be further improved by interpreting

the similarity

distance between words; for example,

love

, lovwww, loveee and luv as one entity

’love

Current development in sentiment analysisSlide23

Analyzing e-learning blogs and reviews can help in providing better services to the users and improve

the teaching

-learning processJensen et al. proposed a technique by which about 150,000 twitter messages were analyzed. The results obtained conveyed that 19% mentioned a brand name, and 20% expressed sentiments about brands, among which about 50% spoke positively and 33% spoke negatively. Current development in sentiment analysisSlide24

Extensive research has been carried out in the field of sentiment analysais

- text sentiment

classifiers, effect analysis, automatic survey analysis, opinion extraction, or recommender systems In this paper, they have presented different approaches available to analyze sentiment at different levels. Based on the needs of the data to

be analyzed

, a particular approach can be chosen. For example, to analyze reviews about a mobile, feature-level sentiment analysis can be carried out. This will help in knowing user’s opinion with respect to various

features

conclusionSlide25

Applying data mining techniques on e-learning reviews and studying e-learning blogs are some of the

challenges faced

in improving the accuracy of the proposed system further.Sentiment analysis of twitter messages can help in making financial, marketing, political decisions. People use tweets to express their opinion about something. They plan to design and develop a system for detecting and visualizing sentiment bias in online articles

The proposed system will be able to dynamically summarize the sentiment for different subtopics and for different websites.

They plan to construct a model which can automatically calculate credibility scores for articles based on sentiment difference between subtopics and between websites.Future workSlide26

B. Pang and L. Lee, “Opinion Mining and Sentiment Analysis”,Foundations

and Trends in Information Retrieval 2(1-2), 2008Hogenboom, A.; van Iterson, P.; Heerschop, B.; Frasincar, F.; Kaymak, U.; , "Determining negation scope and strength in sentiment analysis," Systems, Man, and Cybernetics (SMC), 2011 IEEE International Conference on , vol., no., pp.2589-2594, 9-12 Oct. 2011Chunxi Liu; Li Su; Qingming Huang;

Shuqiang

Jiang; , "News video story sentiment classification and ranking," Multimedia and Expo (ICME), 2011 IEEE International Conference on , vol., no., pp.1-6, 11-15 July 2011Hajmohammadi, M., Ibrahim, R., Ali Othman, Z.. Opinion

Mining and

Sentiment Analysis: A Survey. International Journal

of Computers

& Technology, North America, 2, jun. 2012

referencesSlide27

THANK YOU