Clustering Sentences with Density Peaks for Multi-document Summarization

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1262–1267, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Clustering Sentences with Density Peaks for Multi-document Summarization Yang Zhang Shenzhen Graduate School Peking University, China ecezhangy@sz.pku.edu.cn Yunqing Xia Dept. of Comp. Sci. and Tech. Tsinghua University, China yqxia@tsinghua.edu.cn Yi Liu IMSL, PKU-HKUST Shenzhen Hong Kong Institution, China yi.liu@imsl.org.cn Wenmin Wang Shenzhen Graduate School Peking University, China wangwm@ece.pku.edu.cn Abstract Multi-document Summarization (MDS) is of great value to many real world applications. Many scoring models are proposed to select appropriate sentences from documents to form the summary, in which the clustering-based methods are popular. In this work, we propose a unified sentence scoring model which mea- sures representativeness and diversity at the same time. Experimental results on DUC04 demonstrate that our MDS method outper- forms the DUC04 best method and the ex- isting clustering-based methods, and it yields close results compared to the state-of-the-art generic MDS methods. Advantages of the proposed MDS method are two-fold: (1) The density peaks clustering algorithm is firstly adopted, which is effective and fast. (2) No external resources such as Wordnet and Wikipedia or complex language parsing al- gorithms is used, making reproduction and deployment very easy in real environment. 1 Introduction Document summarization is the process of gener- ating a generic or topic-focused summary by re- ducing documents in size while retaining the main characteristics of original documents(Wang et al., 2011). The summary may be formed in a variety of different ways, which are generally categorized as abstractive and extractive(Shen et al., 2007). In this paper, we address the problem of generic multi- document summarization (MDS). An effective sum- marization method should properly consider the fol- lowing three important issues: representativeness, diversity, conciseness. Many scoring models are proposed to select ap- propriate sentences from documents to form the summary, in which the clustering-based methods are popular. Some researchers address the sentence scoring task in an isolation manner(Radev et al., 2004; Wang et al., 2008; Wan and Yang, 2008) (i.e., clustering and ranking are two independent steps). Others handle the sentence ranking task in a mutuality manner(Cai and Li, 2013; Cai et al., 2010; Wang et al., 2011) (i.e., clustering improves ranking and vice versa). Two drawbacks of the existing clustering-based methods are worth noting. First, extra algorithms are required to determine the number of clusters beforehand. Second, models are required to rank or score sentences within and across the clusters after clustering. Our proposed MDS method is inspired by the recent work on density peaks clustering (DPC) al- gorithm published on Science (Rodriguez and Laio, 2014). The underlying assumption is that clus- ter centers are characterized by a higher density than their neighbors and by a relatively large dis- tance from points with higher densities. In this paper, we adapt the density peaks clustering algo- rithm(Rodriguez and Laio, 2014) to simultaneously cluster sentences and rank them in the mutuality manner. Thanks to the density peaks …

Content truncated to 500 words