Query-Focused Multi-Document Summarization Using Co-Training Based Semi-Supervised Learning

Query-Focused Multi-Document Summarization Using Co-Training Based Semi-Supervised Learning ∗∗∗∗ Po Hu a,b, Donghong Ji a, Hai Wang c, and Chong Teng a a 129 Luoyu Road, Computer School, Wuhan University, Wuhan 430079, China phu@mail.ccnu.edu.cn, donghong_ji2000@yahoo.com.cn, tchong616@126.com b 152 Luoyu Road, Department of Computer Science, Huazhong Normal University, Wuhan 430079, China c 463 Guanshan Road, School of Electronics and Information, Wuhan Polytechnic, Wuhan 430074, China whseaking@163.com Abstract. This paper presents a novel approach to query-focused multi-document summarization. As a good biased summary is expected to keep a balance among query relevance, content salience and information diversity, the approach first makes use of both the content feature and the relationship feature to select a number of sentences via the co- training based semi-supervised learning, which can identify the query relevant sentences beyond a single point of view. Then the ranking algorithm based on Markov chain random walks is employed on the relevant sentences by encouraging content salience and information diversity in a unified framework. The final summary focusing on the integration of relevance, salience and diversity is created after several sentences with the highest overall ranking scores are extracted. We performed experiments on DUC2007 dataset and the evaluation results show that the proposed approach can achieve significant improvement over standard baseline approaches and gain comparable performance to the state-of-the-art systems. Keywords: multi-document summarization, co-training based semi-supervised learning, graph based sentence ranking ∗ The work reported in this paper was supported by the Major Research Plan of National Natural Science Foundation of China (90820005, 90920005), National Natural Science Foundation of China (60773011, 60773167) and Wuhan University 985 Project (985yk004). Copyright 2009 by Po Hu, Donghong Ji, Hai Wang, and Chong Teng 1 Introduction Query-focused multi-document summarization has attracted much attention in recent years. Different from generic summarization, it aims to provide more personalized information for a given query. By automatically capturing relevant and salient content from a large amount of searching results and showing them in a concise way, the sort of summarization can aid people to quickly access and digest their interested information. It also provides an effective means to diverse applications such as question answering system, personalized information retrieval, personalized news recommender, etc. To date, the most influential annual evaluation workshop for automatic summarization research is the Document Understanding Conference (DUC or now TAC), which provides a large-scale test benchmark as well as common evaluation procedures for researchers to share their ideas and experiences. 190 23rd Pacific Asia Conference on Language, Information and Computation, pages 190–199 The critical issues in query-focused multi-document summarization are as follows: The first one is that the information contained in the generated summary should be highly related to the given query. The second issue is how to take salience and diversity into account when selecting a batch of sentences or other textual units as representatives for a summary. As the allowed capacity in a summary is usually limited, it will be inappropriate to put all the informative sentences from different documents into the summary for they …

Content truncated to 500 words