Generating Aspect-oriented Multi-Document Summarization with Event-aspect model

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1137–1146, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Generating Aspect-oriented Multi-Document Summarization with Event-aspect model Peng Li1 and Yinglin Wang1 and Wei Gao2and Jing Jiang3 1 Department of Computer Science and Engineering, Shanghai Jiao Tong University 2 Department of Systems Engineering and Engineering Management, Chinese University of Hong Kong 3 School of Information Systems, Singapore Management University {lipeng, ylwang@sjtu.edu.cn} {wgao@se.cuhk.edu.hk} {jingjiang@smu.edu.sg} Abstract In this paper, we propose a novel approach to automatic generation of aspect-oriented sum- maries from multiple documents. We first de- velop an event-aspect LDA model to cluster sentences into aspects. We then use extend- ed LexRank algorithm to rank the sentences in each cluster. We use Integer Linear Pro- gramming for sentence selection. Key features of our method include automatic grouping of semantically related sentences and sentence ranking based on extension of random walk model. Also, we implement a new sentence compression algorithm which use dependency tree instead of parser tree. We compare our method with four baseline methods. Quantita- tive evaluation based on Rouge metric demon- strates the effectiveness and advantages of our method. 1 Introduction In recent years, there has been much interest in the task of multi-document summarization. In this paper, we study the task of automatically generat- ing aspect-oriented summaries from multiple docu- ments. The goal of aspect-oriented summarization is to present the most important content to the us- er in a condensed form and a well-organized struc- ture to satisfy the user’s needs. A summary should follow a readable structure and cover all the aspect- s users are interested in. For example, a summary about natural disasters should include aspects about what happened, when/where it happened, reasons, damages, rescue efforts, etc. and these aspects may be scattered in multiple articles written by different news agencies. Our goal is to automatically collect aspects and construct summaries from multiple doc- uments. Aspect-oriented summarization can be used in many scenarios. First of all, it can be used to gener- ate Wikipedia-like summary articles, especially used to generate introduction sections that summarizes the subject of articles before the table of contents and other elaborate sections. Second, opinionat- ed text often contains multiple viewpoints about an issue generated by different people. Summarizing these multiple opinions can help people easily di- gest them. Furthermore, combined with search en- gines and question&answering systems, we can bet- ter organize the summary content based on aspects to improve user experience. Despite its usefulness, the problem of modeling domain specific aspects for multi-document summa- rization has not been well studied. The most relevant work is by (Haghighi and Vanderwende, 2009) on exploring content models for multi-document sum- marization. They proposed a HIERSUM model for finding the subtopics or aspects which are combined by using KL-divergence criterion for selecting rel- evant sentences. They introduced a general con- tent distribution and several specific content distri- butions to discover the topic and aspects for a s- ingle document …

Content truncated to 500 words