Citation Summarization Through Keyphrase Extraction

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 895–903, Beijing, August 2010 Citation Summarization Through Keyphrase Extraction Vahed Qazvinian Department of EECS University of Michigan vahed@umich.edu Dragomir R. Radev School of Information and Department of EECS University of Michigan radev@umich.edu Arzucan ¨Ozg¨ur Department of EECS University of Michigan ozgur@umich.edu Abstract This paper presents an approach to sum- marize single scientific papers, by extract- ing its contributions from the set of cita- tion sentences written in other papers. Our methodology is based on extracting sig- nificant keyphrases from the set of cita- tion sentences and using these keyphrases to build the summary. Comparisons show how this methodology excels at the task of single paper summarization, and how it out-performs other multi-document sum- marization methods. 1 Introduction In recent years statistical physicists and computer scientists have shown great interest in analyzing complex adaptive systems. The study of such sys- tems can provide valuable insight on the behav- ioral aspects of the involved agents with potential applications in economics and science. One such aspect is to understand what motivates people to provide the n+1st review of an artifact given that they are unlikely to add something significant that has not already been said or emphasized. Cita- tions are part of such complex systems where ar- ticles use citations as a way to mention different contributions of other papers, resulting in a col- lective system. The focus of this work is on the corpora cre- ated based on citation sentences. A citation sen- tence is a sentence in an article containing a ci- tation and can contain zero or more nuggets (i.e., non-overlapping contributions) about the cited ar- ticle. For example the following sentences are a few citation sentences that appeared in the NLP literature in past that talk about Resnik’s work. The STRAND system (Resnik, 1999), for example, uses structural markup information from the pages, without looking at their content, to attempt to align them. Resnik (1999) addressed the issue language identification for finding Web pages inof the languages of interest. Mining the Web for bilingual text (Resnik, 1999) is not likely to provide sufficient quantities of high quality data.. The set of citations is important to analyze be- cause human summarizers have put their effort collectively but independently to read the target article and cite its important contributions. This has been shown in other work too (Elkiss et al., 2008; Nanba et al., 2004; Qazvinian and Radev, 2008; Mei and Zhai, 2008; Mohammad et al., 2009). In this work, we introduce a technique to summarize the set of citation sentences and cover the major contributions of the target paper. Our methodology first finds the set of keyphrases that represent important information units (i.e., nuggets), and then finds the best set of k sentences to cover more, and more important nuggets. Our results confirm the effectiveness of the method and show that it outperforms other state of the art summarization techniques. Moreover, as shown in the paper, this method …

Content truncated to 500 words