An overview of concepts of Sentiment Analysis

sentiment Analysis

1. Sentiment Analysis
An Overview of Concepts and
Selected Techniques

2. Terms
Sentiment
A thought, view, or attitude, especially one
based mainly on emotion instead of reason
Sentiment Analysis
 aka opinion mining
use of natural language processing (NLP) and
computational techniques to automate the
extraction or classification of sentiment from
typically unstructured text

3. Motivation
Consumer information
Product reviews
Marketing
Consumer attitudes
Trends
Politics
Politicians want to know voters’ views
Voters want to know policitians’ stances and who else
supports them
Social
Find like-minded individuals or communities

4. Problem
Which features to use?
Words (unigrams)
Phrases/n-grams
Sentences
How to interpret features for sentiment
detection?
Bag of words (IR)
Annotated lexicons (WordNet, SentiWordNet)
Syntactic patterns
Paragraph structure

5. Challenges
Harder than topical classification, with
which bag of words features perform well
Must consider other features due to…
Subtlety of sentiment expression
irony
expression of sentiment using neutral words
Domain/context dependence
words/phrases can mean different things in different
contexts and domains
Effect of syntax on semantics

6. Approaches
Machine learning
Naïve Bayes
Maximum Entropy Classifier
SVM
Markov Blanket Classifier
Accounts for conditional feature dependencies
Allowed reduction of discriminating features from
thousands of words to about 20 (movie review
domain)
Unsupervised methods
Use lexicons
Assume pairwise
independent features

7. LingPipe Polarity Classifier
First eliminate objective sentences, then
use remaining sentences to classify
document polarity (reduce noise)

8. LingPipe Polarity Classifier
Uses unigram features extracted from
movie review data
Assumes that adjacent sentences are
likely to have similar subjective-objective
(SO) polarity
Uses a min-cut algorithm to efficiently
extract subjective sentences

9. LingPipe Polarity Classifier
Graph for classifying three items.

10. LingPipe Polarity Classifier
Accurate as baseline but uses only 22% of
content in test data (average)
Metrics suggests properties of movie
review structure

11. SentiWordNet
Based on WordNet “synsets”
http://wordnet.princeton.edu/
Ternary classifier
Positive, negative, and neutral scores for each
synset
Provides means of gauging sentiment for
a text

12. SentiWordNet: Construction
Created training sets of synsets, L
p and L
n
Start with small number of synsets with fundamentally
positive or negative semantics, e.g., “nice” and “nasty”
Use WordNet relations, e.g., direct antonymy, similarity,
derived-from, to expand L
p
and L
n
over K iterations
L
o
(objective) is set of synsets not in L
p
or L
n
Trained classifiers on training set
Rocchio and SVM
Use four values of K to create eight classifiers with
different precision/recall characteristics
As K increases, P decreases and R increases

13. SentiWordNet: Results
24.6% synsets with Objective<1.0
Many terms are classified with some degree of
subjectivity
10.45% with Objective<=0.5
0.56% with Objective<=0.125
Only a few terms are classified as definitively
subjective
Difficult (if not impossible) to accurately
assess performance

14. SentiWordNet: How to use it
Use score to select features (+/-)
e.g. Zhang and Zhang (2006) used words in
corpus with subjectivity score of 0.5 or greater
Combine pos/neg/objective scores to
calculate document-level score
e.g. Devitt and Ahmad (2007) conflated
polarity scores with a Wordnet-based graph
representation of documents to create
predictive metrics

15. References
1.http://www.answers.com/sentiment , 9/22/08
B. Pang, L. Lee, and S. Vaithyanathan, “Thumbs up? Sentiment
classification using machine learning techniques,” in Proc Conf
on Empirical Methods in Natural Language Processing (EMNLP) ,
pp. 79–86, 2002.
Esuli A, Sebastiani F. SentiWordNet: A Publicly Available Lexical
Resource for Opinion Mining. In: Proc of LREC 2006 - 5th Conf
on Language Resources and Evaluation, 2006.
Zhang E, Zhang Y. UCSC on TREC 2006 Blog Opinion Mining.
TREC 2006 Blog Track, Opinion Retrieval Task.
Devitt A, Ahmad K.
 Sentiment Polarity Identification in Financial
News: A Cohesion-based Approach . ACL 2007.
Bo Pang , Lillian Lee, A sentimental education: sentiment
analysis using subjectivity summarization based on minimum
cuts, Proceedings of the 42nd Annual Meeting on Association for
Computational Linguistics, p.271-es, July 21-26, 2004.