tailieunhanh - Báo cáo khoa học: "Extractive Summarization Based on Event Term Clustering"

Most existing event-based summarization approaches rely on the statistical features derived from documents and generally associated with Event-based summarization extracts and single events, but they neglect the relations among organizes summary sentences in terms of events. However, events are commonly related the events that the sentences describe. In with one another especially when the documents to this work, we focus on semantic relations be summarized are about the same or very similar among event terms. . | Extractive Summarization Based on Event Term Clustering Maofu Liu1 2 Wenjie Li1 Mingli Wu1 and Qin Lu1 department of Computing 2College of Computer Science and Technology The Hong Kong Polytechnic University csmfliu cswjli csmlwu csluqin @ Abstract Event-based summarization extracts and organizes summary sentences in terms of the events that the sentences describe. In this work we focus on semantic relations among event terms. By connecting terms with relations we build up event term graph upon which relevant terms are grouped into clusters. We assume that each cluster represents a topic of documents. Then two summarization strategies are investigated . selecting one term as the representative of each topic so as to cover all the topics or selecting all terms in one most significant topic so as to highlight the relevant information related to this topic. The selected terms are then responsible to pick out the most appropriate sentences describing them. The evaluation of clustering-based summarization on DUC 2001 document sets shows encouraging improvement over the well-known PageRank-based summarization. 1 Introduction Event-based extractive summarization has emerged recently Filatova and Hatzivassiloglou 2004 . It extracts and organizes summary sentences in terms of the events that sentences describe. We follow the common agreement that event can be formulated as Who did What to Whom When and Where and did What denotes the key element of an event . the action within the formulation. We approximately define the verbs and action nouns as the event terms which can characterize or partially characterize the event occurrences. Wuhan University of Science and Technology mfliuchina@ Most existing event-based summarization approaches rely on the statistical features derived from documents and generally associated with single events but they neglect the relations among events. However events are commonly related with one another especially