Identifying Sources of Opinions with Conditional Random Fields and Extraction Patterns

Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 355–362, Vancouver, October 2005. c⃝2005 Association for Computational Linguistics Identifying Sources of Opinions with Conditional Random Fields and Extraction Patterns Yejin Choi and Claire Cardie Department of Computer Science Cornell University Ithaca, NY 14853 {ychoi,cardie}@cs.cornell.edu Ellen Riloff and Siddharth Patwardhan School of Computing University of Utah Salt Lake City, UT 84112 {riloff,sidd}@cs.utah.edu Abstract Recent systems have been developed for sentiment classification, opinion recogni- tion, and opinion analysis (e.g., detect- ing polarity and strength). We pursue an- other aspect of opinion analysis: identi- fying the sources of opinions, emotions, and sentiments. We view this problem as an information extraction task and adopt a hybrid approach that combines Con- ditional Random Fields (Lafferty et al., 2001) and a variation of AutoSlog (Riloff, 1996a). While CRFs model source iden- tification as a sequence tagging task, Au- toSlog learns extraction patterns. Our re- sults show that the combination of these two methods performs better than either one alone. The resulting system identifies opinion sources with 79.3% precision and 59.5% recall using a head noun matching measure, and 81.2% precision and 60.6% recall using an overlap measure. 1 Introduction In recent years, there has been a great deal of in- terest in methods for automatically identifying opin- ions, emotions, and sentiments in text. Much of this research explores sentiment classification, a text categorization task in which the goal is to classify a document as having positive or negative polar- ity (e.g., Das and Chen (2001), Pang et al. (2002), Turney (2002), Dave et al. (2003), Pang and Lee (2004)). Other research efforts analyze opinion ex- pressions at the sentence level or below to recog- nize opinions, their polarity, and their strength (e.g., Dave et al. (2003), Pang and Lee (2004), Wilson et al. (2004), Yu and Hatzivassiloglou (2003), Wiebe and Riloff (2005)). Many applications could ben- efit from these opinion analyzers, including prod- uct reputation tracking (e.g., Morinaga et al. (2002), Yi et al. (2003)), opinion-oriented summarization (e.g., Cardie et al. (2004)), and question answering (e.g., Bethard et al. (2004), Yu and Hatzivassiloglou (2003)). We focus here on another aspect of opinion analysis: automatically identifying the sources of the opinions. Identifying opinion sources will be especially critical for opinion-oriented question- answering systems (e.g., systems that answer ques- tions of the form “How does [X] feel about [Y]?”) and opinion-oriented summarization systems, both of which need to distinguish the opinions of one source from those of another.1 The goal of our research is to identify direct and indirect sources of opinions, emotions, sentiments, and other private states that are expressed in text. To illustrate the nature of this problem, consider the examples below: S1: Taiwan-born voters favoring independence... 1In related work, we investigate methods to identify the opinion expressions (e.g., Riloff and Wiebe (2003), Wiebe and Riloff (2005), Wilson et al. (2005)) and the nesting structure of sources (e.g., Breck and Cardie (2004)). The target of each opinion, i.e., …

Content truncated to 500 words