Affect analysis in text and speech

Affect analysis in text and speech
CIS630 Spring 2013
Lecture 2
Course objectives
 Overview theories and approaches to affect and opinion
analysis in text and speech
 Personalized project to allow students to deepen their
understanding and pursue applications outside the course
 Learn to comfortably discuss academic papers, with
critiques, summaries and discussions
 Improve written and oral presentation skills
Grade breakdown
 [30%] Project 1: text-based analysis on NYT
 [40%] Project 2: student proposed project
 Proposal
 Progress presentation (in class)
 Final report (written)
 [15%] Questions on assigned reading and class participation
 [15%] Leading a discussion
Organization: assigned readings
 Readings assigned for each class
 Students should do the readings BEFORE class starting next
 Submit three questions related to the assigned reading
 Due on by 9am the day of class
 Discussion leaders can use these to prepare their presentation
Project 1
 General theme: Affect analysis in the NYT
 Project presentations: Feb 25/27
 Project report due March 6
 Will discuss more specifics in the second half of class today
Project 2
 One page proposals due March 1
 Students are encouraged to integrate the topic with other
research or project that they are doing
 Proposal presentation April 1 (~10 mins)
 Preliminary results April 17/22 (~20mins)
 Final projects due: May 1
 Teams of two may work together on the project
Reading 1
 Sentiment analysis and opinion mining, Bing Liu
 Jan 14 (pp.1 –37)
 Jan 16 (pp. 37—79)
 Jan 21 MLK day---no class
 Jan 23 (pp. 79—134)
 After that we will start with two readings per class, questions
and student-led discussions
 Opinions are central to almost all human activities and are
key influencers of our behavior
 What we pay attention to is determined by affect
 What we remember is modulated by affect. Stress impedes
memory, positive attitude strengthens it
 May be expressed rather differently cross-culturally
 Opinions and related concepts such as sentiments, evaluations,
attitudes, and emotions are the subjects of study of sentiment
analysis and opinion mining
What others think
 Whenever we need to make a decision, we want to know
others’ opinion
 Is it the opinions of others, or is it facts that may help us form
our opinion that we are looking for?
 Company/product image analysis
 Using sentiment to predict sales
 Using reviews to rank products and merchants
 Link between public opinion in blogs and Twitter and NFL betting
Twitter sentiment to predict election results
Analysis of political standpoints
Predicting comment volumes of political blogs
Twitter data, movie reviews and blogs to predict movie box-office
 Analyze sentiment flow in social networks
 Twitter moods to predict the stock market, trading
strategies, particular stocks
 Assessment of global happiness based on Facebook posts
Levels of analysis
Document level: assume the document talks about a single
entity, predict the sentiment
Sentence level: Determine if a sentence expresses a positive,
negative or neutral opinion
Subjective (express internal speaker states that cannot be
verified) vs objective sentence (expressing facts)
“We bought the car last month and the windshield wiper has
fallen off ”.
 objective sentence that expresses negative opinion
Entity and aspect level
Example from Google products
Posted by: John Smith
Date: Sep 10, 2011
(1) I bought a Canon G12 camera six months ago (2) I simply
love it. (3) The picture quality is amazing. (4) The battery life
is also long. (5) However, my wife thinks it is too heavy for
Opinion (definition)
 An opinion is a quintuple
(ei ,aij ,sijkl,hk ,t l )
e: entity (nouns/NPs)
a: aspect (nouns/NPs)
s: positive, negative or neutral sentiment (explicit or implicit)
h: opinion holder
t: time
Document sentiment classification
GENERAL aspect: covers the entity as a whole
Binary classification: positive or negative
Regression: scale, number of starts
a standard baseline for such tasks is the use of the individual words
in the text
-- terms and their frequency
-- part of speech
-- sentiment words and phrases
-- sentiment shifters (hardly, mostly, didn’t)
-- syntactic dependencies
Parts of speech and sentiment
 Verbs: can evoke/convey sentiment by association of the
results of the action
 Break, Lose, Win, Murder
 Nouns: usually the opinion/sentiment targets; some can be
 Adjectives and adverbs: by definition convey properties, so if
the properties are worth mentioning probably has some
affective value
Project 1 proposal
 Use existing sentiment resources to rank texts in the NYT in
terms of their opinion and affect content
 Find words that are distributed differently on the top and
bottom ranked documents
 Find nouns that are modified by adjectives
 JJ NN, NN is JJ
 Rank entities by the diversity of qualifications and the ratio of
times they are mentioned with and without adjective
 Similarly with verbs and adverbs

similar documents