Đang chuẩn bị liên kết để tải về tài liệu:
Báo cáo khoa học: "Co-Feedback Ranking for Query-Focused Summarization"

In this paper, we propose a novel ranking framework – Co-Feedback Ranking (CoFRank), which allows two base rankers to supervise each other during the ranking process by providing their own ranking results as feedback to the other parties so as to boost the ranking performance. The mutual ranking refinement process continues until the two base rankers cannot learn from each other any more. The overall performance is improved by the enhancement of the base rankers through the mutual learning mechanism. We apply this framework to the sentence ranking problem in query-focused summarization and evaluate its effectiveness on the DUC. | Co-Feedback Ranking for Query-Focused Summarization Furu Wei1 2 3 Wenjie Li1 and Yanxiang He2 1 The Hong Kong Polytechnic University Hong Kong 2 Wuhan University China csfwei cswjli @comp.polyu.edu.hk frwei yxhe @whu.edu.cn 3 IBM China Research Laboratory Beijing China Abstract In this paper we propose a novel ranking framework - Co-Feedback Ranking CoFRank which allows two base rankers to supervise each other during the ranking process by providing their own ranking results as feedback to the other parties so as to boost the ranking performance. The mutual ranking refinement process continues until the two base rankers cannot learn from each other any more. The overall performance is improved by the enhancement of the base rankers through the mutual learning mechanism. We apply this framework to the sentence ranking problem in query-focused summarization and evaluate its effectiveness on the DUC 2005 data set. The results are promising. 1 Introduction and Background Sentence ranking is the issue of most concern in extractive summarization. Feature-based approaches rank the sentences based on the features elaborately designed to characterize the different aspects of the sentences. They have been extensively investigated in the past due to their easy implementation and the ability to achieve promising results. The use of featurebased ranking has led to many successful e.g. top five systems in DUC 2005-2007 query-focused summarization Over et al. 2007 . A variety of statistical and linguistic features such as term distribution sentence length sentence position and named entity etc. can be found in literature. Among them query relevance centroid Radev et al. 2004 and signature term Lin and Hovy 2000 are most remarkable. There are two alternative approaches to integrate the features. One is to combine features into a unified representation first and then use it to rank the sentences. The other is to utilize rank fusion or rank aggregation techniques to combine the .