Generating domain-specific clues using news corpus for sentiment classification

Abstract

The domain-dependent nature of sentiment analysis calls for domain-specific sentiment clues. This paper addresses the problem of automatically generating such domain-specific sentiment clues and proposes an effective solution. The main idea is to bootstrap from a small seed set and generate new clues by using syntactic dependency and collocation information between sentiment clues and sentence-level topics that are defined to be a primary subject of a sentiment expression (e.g., event, company, and person). Our experiments show that the automatically extracted clues are effective for sentiment classification. Copyright © 2010, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.