An Exploration of Document Impact on Graph-Based Multi-Document Summarization

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 755–762, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics An Exploration of Document Impact on Graph-Based Multi-Document Summarization Xiaojun Wan Institute of Compute Science and Technology Peking University Beijing 100871, China wanxiaojun@icst.pku.edu.cn Abstract The graph-based ranking algorithm has been recently exploited for multi-document sum- marization by making only use of the sen- tence-to-sentence relationships in the documents, under the assumption that all the sentences are indistinguishable. However, given a document set to be summarized, dif- ferent documents are usually not equally im- portant, and moreover, different sentences in a specific document are usually differently im- portant. This paper aims to explore document impact on summarization performance. We propose a document-based graph model to in- corporate the document-level information and the sentence-to-document relationship into the graph-based ranking process. Various meth- ods are employed to evaluate the two factors. Experimental results on the DUC2001 and DUC2002 datasets demonstrate that the good effectiveness of the proposed model. More- over, the results show the robustness of the proposed model. 1 Introduction Multi-document summarization aims to produce a summary describing the main topic in a document set, without any prior knowledge. Multi-document summary can be used to facilitate users to quickly understand a document cluster. For example, a number of news services (e.g. NewsInEssence 1) have been developed to group news articles into news topics, and then produce a short summary for each news topic. Users can easily understand the topic they have interest in by taking a look at the short summary, without looking into each individ- ual article within the topic cluster. 1http://lada.si.umich.edu:8080/clair/nie1/nie.cgi Automated multi-document summarization has drawn much attention in recent years. In the com- munities of natural language processing and infor- mation retrieval, a series of workshops and conferences on automatic text summarization (e.g. NTCIR, DUC), special topic sessions in ACL, COLING, and SIGIR have advanced the summari- zation techniques and produced a couple of ex- perimental online systems. A particular challenge for multi-document sum- marization is that a document set might contain diverse information, which is either related or un- related to the main topic, and hence we need effec- tive summarization methods to analyze the information stored in different documents and ex- tract the globally important information to reflect the main topic. In recent years, both unsupervised and supervised methods have been proposed to analyze the information contained in a document set and extract highly salient sentences into the summary, based on syntactic or statistical features. Most recently, the graph-based models have been successfully applied for multi-document summarization by making use of the “voting” or “recommendations” between sentences in the documents (Erkan and Radev, 2004; Mihalcea and Tarau, 2005; Wan and Yang, 2006). The model first constructs a directed or undirected graph to reflect the relationships between the sentences and then applies the graph-based ranking algorithm to compute the rank scores for the sentences. The sentences with large rank scores are chosen into the summary. However, …

Content truncated to 500 words