Determining the Sentiment of Opinions

Determining the Sentiment of Opinions Soo-Min Kim Information Sciences Institute University of Southern California 4676 Admiralty Way Marina del Rey, CA 90292-6695 skim@isi.edu Eduard Hovy Information Sciences Institute University of Southern California 4676 Admiralty Way Marina del Rey, CA 90292-6695 hovy@isi.edu Abstract Identifying sentiments (the affective parts of opinions) is a challenging problem. We present a system that, given a topic, automatically finds the people who hold opinions about that topic and the sentiment of each opinion. The system contains a module for determining word sentiment and another for combining sentiments within a sentence. We experiment with various models of classifying and combining sentiment at word and sentence levels, with promising results. 1 Introduction What is an opinion? The many opinions on opinions are reflected in a considerable literature (Aristotle 1954; Perelman 1970; Toulmin et al. 1979; Wallace 1975; Toulmin 2003). Recent computational work either focuses on sentence ‘subjectivity’ (Wiebe et al. 2002; Riloff et al. 2003), concentrates just on explicit statements of evaluation, such as of films (Turney 2002; Pang et al. 2002), or focuses on just one aspect of opinion, e.g., (Hatzivassiloglou and McKeown 1997) on adjectives. We wish to study opinion in general; our work most closely resembles that of (Yu and Hatzivassiloglou 2003). Since an analytic definition of opinion is probably impossible anyway, we will not summarize past discussion or try to define formally what is and what is not an opinion. For our purposes, we describe an opinion as a quadruple [Topic, Holder, Claim, Sentiment] in which the Holder believes a Claim about the Topic, and in many cases associates a Sentiment, such as good or bad, with the belief. For example, the following opinions contain Claims but no Sentiments: “I believe the world is flat” “The Gap is likely to go bankrupt” “Bin Laden is hiding in Pakistan” “Water always flushes anti-clockwise in the southern hemisphere” Like Yu and Hatzivassiloglou (2003), we want to automatically identify Sentiments, which in this work we define as an explicit or implicit expression in text of the Holder’s positive, negative, or neutral regard toward the Claim about the Topic. (Other sentiments we plan to study later.) Sentiments always involve the Holder’s emotions or desires, and may be present explicitly or only implicitly: “I think that attacking Iraq would put the US in a difficult position” (implicit) “The US attack on Iraq is wrong” (explicit) “I like Ike” (explicit) “We should decrease our dependence on oil” (implicit) “Reps. Tom Petri and William F. Goodling asserted that counting illegal aliens violates citizens’ basic right to equal representation” (implicit) In this paper we address the following challenge problem. Given a Topic (e.g., “Should abortion be banned?”) and a set of texts about the topic, find the Sentiments expressed about (claims about) the Topic (but not its supporting subtopics) in each text, and identify the people who hold each sentiment. To avoid the problem of differentiating between shades of sentiments, we simplify the problem to: identify just expressions of positive, negative, or …

Content truncated to 500 words