Keynote

Speaker: (Information Sciences Institute, University of Southern California)

What are Subjectivity, Sentiment and Affect?

Abstract: Pragmatics —the aspects of text that signal interpersonal and situational information, complementing semantics— has been almost totally ignored in Natural Language Processing. But in the past five to eight years there has been a surge of research on the general topic of ‘opinion’, also called ‘sentiment’. Generally, research focuses on the determining the author’s opinion/sentiment about some topic within a given fragment of text. Since opinions may differ, it is granted that the author’s opinion is ‘subjective’, and the effectiveness of an opinion-determination system is measured by comparing against a gold-standard set of human annotations.
But what does ‘subjectivity’ actually mean? What are ‘opinion’ and ‘sentiment’? Lately, researchers are also starting to talk about ‘affect’, and even ‘emotion’. What are these notions, and how do they differ from one another?
Unfortunately, a survey of the research done to date shows a disturbing lack of clarity on these questions. Very few papers bother to define their terms, but simply take a set of valences such as Good–Neutral–Bad to be sufficient. More recent work acknowledges the need to specify what the opinion actually applies to, and attempts also to determine the theme. Lately, several identify the holder of the opinion. Some even try to estimate the strength of the expressed opinion.
The trouble is, the same aspect of the same object can be considered Good by one person and Bad by another, and we can often understand both their points of view. There is much more to opinion/sentiment than simply matching words and phrases that attach to the theme, and computing a polarity score. People give reasons why they like or dislike something, and these reasons pertain to their goals and plans in the case of opinions) or their deeper emotional states (in the case of affect).
In this talk I outline a model of sentiment/opinion and of affect, and show that they appear in text in a fairly structured way, with various components. I show how proper understanding requires the reader to build some kind of person profile of the author, and claim that for systems to do adequate understanding of sentiments, opinions, and affects, they will need to do so as well. This is not a trivial challenge, and it opens the door to a whole new line of research with many fascinating and practical aspects.

Program Schedule


08:45 - 9:00 Opening Remarks
09:00 - 10:00 Keynote: What are Subjectivity, Sentiment, and Affect?
Eduard Hovy
10:00 - 10:30 Coffee/Tea Break

Session 1:


10:30 - 10:50 On Using Twitter to Monitor Political Sentiment and Predict Election Results
Adam Bermingham and Alan Smeaton
10:50 - 11:10 Taking Refuge in Your Personal Sentic Corner
Erik Cambria, Amir Hussain and Chris Eckl
11:10 - 11:30 Towards automatic detection of antisocial behavior from texts
Myriam Munezero, Tuomo Kakkonen and Calkin Montero
11:30 - 11:45 User Profile Construction in the TWIN Personality-based Recommender System
Alexandra Roshchina, John Cardiff and Paolo Rosso
11:45 - 12:00 Enriching Social Communication through Semantics and Sentics
Praphul Chandra, Erik Cambria and Alvin Pradeep
12:00 - 14:00 Lunch

Session 2:


14:00 - 14:20 Emotion Modeling from Writer/Reader Perspectives Using a Microblog Dataset
Yi-jie Tang and Hsin-Hsi Chen
14:20 - 14:35 Introducing Argumention in Opinion Analysis: Language and Reasoning Challenges
Leila Amgoud, Florence Bannay, Charlotte Costedoat, Patrick Saint-Dizier and Camille Albert
14:35 - 14:55 Sense-level Subjectivity in a Multilingual Setting
Carmen Banea, Rada Mihalcea and Janyce Wiebe
14:55 - 15:15 Applying Sentiment-oriented Sentence Filtering to Multilingual Review Classification
Takashi Inui and Mikio Yamamoto
15:15 - 15:30 Chinese Sentiment Analysis Using Maximum Entropy
Huey Yee Lee and Hemnaath Renganathan
15:30 - 16:00 Coffee/Tea Break

Session 3:


16:00 - 16:20 Analyzing Emotional Statements - Roles of General and Physiological Variables
Dipankar Das and Sivaji Bandyopadhyay
16:20 - 16:35 Incorporating Lexicon Knowledge into SVM Learning to Improve Sentiment Classification
Ji Fang and Bi Chen
16:35 - 16:55 What is new? News media, General Elections, Sentiment, and Named Entities
Khurshid Ahmad, Nicholas Daly and Vanessa Liston
16:55 - 17:15 Towards Enhanced Opinion Classification using NLP Techniques.
Akshat Bakliwal, Piyush Arora, Ankit Patil and Vasudeva Varma