A Sentence Compression Based Framework to Query-Focused Multi-Document Summarization

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1384–1394, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics A Sentence Compression Based Framework to Query-Focused Multi-Document Summarization Lu Wang1 Hema Raghavan2 Vittorio Castelli2 Radu Florian2 Claire Cardie1 1Department of Computer Science, Cornell University, Ithaca, NY 14853, USA {luwang, cardie}@cs.cornell.edu 2IBM T. J. Watson Research Center, Yorktown Heights, NY 10598, USA {hraghav, vittorio, raduf}@us.ibm.com Abstract We consider the problem of using sentence compression techniques to facilitate query- focused multi-document summarization. We present a sentence-compression-based frame- work for the task, and design a series of learning-based compression models built on parse trees. An innovative beam search de- coder is proposed to efficiently find highly probable compressions. Under this frame- work, we show how to integrate various in- dicative metrics such as linguistic motivation and query relevance into the compression pro- cess by deriving a novel formulation of a com- pression scoring function. Our best model achieves statistically significant improvement over the state-of-the-art systems on several metrics (e.g. 8.0% and 5.4% improvements in ROUGE-2 respectively) for the DUC 2006 and 2007 summarization task. 1 Introduction The explosion of the Internet clearly warrants the development of techniques for organizing and presenting information to users in an effective way. Query-focused multi-document summariza- tion (MDS) methods have been proposed as one such technique and have attracted significant at- tention in recent years. The goal of query-focused MDS is to synthesize a brief (often fixed-length) and well-organized summary from a set of topic- related documents that answer a complex ques- tion or address a topic statement. The result- ing summaries, in turn, can support a number of information analysis applications including open- ended question answering, recommender systems, and summarization of search engine results. As further evidence of its importance, the Document Understanding Conference (DUC) has used query- focused MDS as its main task since 2004 to foster new research on automatic summarization in the context of users’ needs. To date, most top-performing systems for multi-document summarization—whether query- specific or not—remain largely extractive: their summaries are comprised exclusively of sen- tences selected directly from the documents to be summarized (Erkan and Radev, 2004; Haghighi and Vanderwende, 2009; Celikyilmaz and Hakkani-T¨ur, 2011). Despite their simplicity, extractive approaches have some disadvantages. First, lengthy sentences that are partly relevant are either excluded from the summary or (if se- lected) can block the selection of other important sentences, due to summary length constraints. In addition, when people write summaries, they tend to abstract the content and seldom use entire sentences taken verbatim from the original documents. In news articles, for example, most sentences are lengthy and contain both potentially useful information for a summary as well as un- necessary details that are better omitted. Consider the following DUC query as input for a MDS system:1 “In what ways have stolen artworks been recovered? How often are suspects arrested or prosecuted for the thefts?” One manually gen- erated summary includes the following sentence but removes the bracketed …

Content truncated to 500 words