A study of fine-grained sentence-level emotion tagging

While there has been much work on sentiment analysis, emotion tagging has not been very well studied. Existing work has generally treated each text article as a unit for emotion tagging. In this work, we argue that it is more useful to perform emotion tagging at the sentence-level and use Conditional Random Fields (CRF) to tag sentences with five emotion tags. We propose and study multiple features, including both basic features defined on a single sentence and dependency features defined on the context of a sentence. We create two test sets, one with email messages and one with product reviews, to evaluate the proposed features. Experimental results show that in general, dependency features are beneficial, and in particular, using relative position features can significantly improve the accuracy. We also present clustering of users based on their emotional profiles as a possible application of sentence-level emotion tagging.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2013

Author dc:creator

• Gurmeet, Singh

Contributors dc:contributor

• Zhai, ChengXiang

• Opinion mining
• Conditional random fields
• Emotion tagging
• Emotional profile

Statement dc:rights

• Copyright 2013 Gurmeet Singh

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/44146

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/44146