Citation-Enhanced Keyphrase Extraction from Research Papers: A Supervised Approach

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1435–1446, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Citation-Enhanced Keyphrase Extraction from Research Papers: A Supervised Approach Cornelia Caragea1, Florin Bulgarov1, Andreea Godea1, Sujatha Das Gollapalli2 1Computer Science and Engineering, University of North Texas, TX, USA 2Institute for Infocomm Research, A*STAR, Singapore ccaragea@unt.edu, FlorinBulgarov@my.unt.edu, AndreeaGodea@my.unt.edu, gsdas@cse.psu.edu Abstract Given the large amounts of online textual documents available these days, e.g., news articles, weblogs, and scientific papers, ef- fective methods for extracting keyphrases, which provide a high-level topic descrip- tion of a document, are greatly needed. In this paper, we propose a supervised model for keyphrase extraction from research pa- pers, which are embedded in citation net- works. To this end, we design novel fea- tures based on citation network informa- tion and use them in conjunction with tra- ditional features for keyphrase extraction to obtain remarkable improvements in per- formance over strong baselines. 1 Introduction Keyphrase extraction is the problem of automat- ically extracting important phrases or concepts (i.e., the essence) of a document. Keyphrases provide a high-level topic description of a docu- ment and are shown to be rich sources of informa- tion for many applications such as document clas- sification, clustering, recommendation, indexing, searching, and summarization (Jones and Stave- ley, 1999; Zha, 2002; Hammouda et al., 2005; Pudota et al., 2010; Turney, 2003). Despite the fact that keyphrase extraction has been widely re- searched in the natural language processing com- munity, its performance is still far from being sat- isfactory (Hasan and Ng, 2014). Many previous approaches to keyphrase extrac- tion generally used only the textual content of a target document to extract keyphrases (Hulth, 2003; Mihalcea and Tarau, 2004; Liu et al., 2010). Recently, Wan and Xiao (2008) proposed a model that incorporates a local neighborhood of a doc- ument. However, their neighborhood is limited to textually-similar documents, where the cosine similarity between the tf-idf vectors of documents is used to compute their similarity. We posit that, in addition to a document’s textual content and textually-similar neighbors, other informative neighborhoods exist that have the potential to im- prove keyphrase extraction. For example, in a scholarly domain, research papers are not isolated. Rather, they are highly inter-connected in giant ci- tation networks, in which papers cite or are cited by other papers. In a citation network, information flows from one paper to another via the citation re- lation (Shi et al., 2010). This information flow and the influence of one paper on another are specifi- cally captured by means of citation contexts, i.e., short text segments surrounding a citation’s men- tion. These contexts are not arbitrary, but they serve as brief summaries of a cited paper. Figure 1 illustrates this idea using a small citation net- work of a paper by Rendle et al. (2010) that cites (Zimdars et al., 2001), (Hu et al., 2008), (Pan and Scholz, 2009) and (Shani et al., 2005) and is cited by (Cheng et al., 2013). …

Content truncated to 500 words