Partially Supervised Coreference Resolution for Opinion Summarization through Structured Rule Learning

Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP 2006), pages 336–344, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Partially Supervised Coreference Resolution for Opinion Summarization through Structured Rule Learning Veselin Stoyanov and Claire Cardie Department of Computer Science Cornell University Ithaca, NY 14850, USA {ves,cardie}@cs.cornell.edu Abstract Combining fine-grained opinion informa- tion to produce opinion summaries is im- portant for sentiment analysis applica- tions. Toward that end, we tackle the problem of source coreference resolution – linking together source mentions that re- fer to the same entity. The partially super- vised nature of the problem leads us to de- fine and approach it as the novel problem of partially supervised clustering. We pro- pose and evaluate a new algorithm for the task of source coreference resolution that outperforms competitive baselines. 1 Introduction Sentiment analysis is concerned with extracting attitudes, opinions, evaluations, and sentiment from text. Work in this area has been motivated by the desire to provide information analysis ap- plications in the arenas of government, business, and politics (e.g. Coglianese (2004)). Addition- ally, sentiment analysis can augment existing NLP applications such as question answering, informa- tion retrieval, summarization, and clustering by providing information about sentiment (e.g. Stoy- anov et al. (2005), Riloff et al. (2005)). To date, research in the area (see Related Work section) has focused on the problem of extracting senti- ment both at the document level (coarse-grained sentiment information), and at the level of sen- tences, clauses, or individual expressions (fine- grained sentiment information). In contrast, our work concerns the summa- rization of fine-grained information about opin- ions. In particular, while recent research ef- forts have shown that fine-grained opinions (e.g. Riloff and Wiebe (2003), Bethard et al. (2004), Wiebe and Riloff (2005)) as well as their sources (e.g. Bethard et al. (2004), Choi et al. (2005), Kim and Hovy (2005)) can be extracted auto- matically, little has been done to create opin- ion summaries, where opinions from the same source/target are combined, statistics are com- puted for each source/target and multiple opinions from the same source to the same target are ag- gregated. A simple opinion summary is shown in figure 1.1 We expect that this type of opinion sum- mary, based on fine-grained opinion information, will be important for information analysis applica- tions in any domain where the analysis of opinions is critical. This paper addresses the problem of opinion summarization by considering the creation of sim- ple opinion summaries like those of figure 1. We propose source coreference resolution — the task of determining which mentions of opinion sources refer to the same entity — as the primary mecha- nism for identifying the set of opinions attributed to each real-world source. For this type of sum- mary, source coreference resolution constitutes an integral step in the process of generating full opin- ion summaries. For example, given the opinion expressions of figure 1, their polarity, and the asso- ciated opinion sources and targets, the bulk of the resulting summary …

Content truncated to 500 words