Automatic Keyphrase Extraction: A Survey of the State of the Art

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1262–1273, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Automatic Keyphrase Extraction: A Survey of the State of the Art Kazi Saidul Hasan and Vincent Ng Human Language Technology Research Institute University of Texas at Dallas Richardson, TX 75083-0688 {saidul,vince}@hlt.utdallas.edu Abstract While automatic keyphrase extraction has been examined extensively, state-of-the- art performance on this task is still much lower than that on many core natural lan- guage processing tasks. We present a sur- vey of the state of the art in automatic keyphrase extraction, examining the major sources of errors made by existing systems and discussing the challenges ahead. 1 Introduction Automatic keyphrase extraction concerns “the au- tomatic selection of important and topical phrases from the body of a document” (Turney, 2000). In other words, its goal is to extract a set of phrases that are related to the main topics discussed in a given document (Tomokiyo and Hurst, 2003; Liu et al., 2009b; Ding et al., 2011; Zhao et al., 2011). Document keyphrases have enabled fast and ac- curate searching for a given document from a large text collection, and have exhibited their potential in improving many natural language processing (NLP) and information retrieval (IR) tasks, such as text summarization (Zhang et al., 2004), text categorization (Hulth and Megyesi, 2006), opin- ion mining (Berend, 2011), and document index- ing (Gutwin et al., 1999). Owing to its importance, automatic keyphrase extraction has received a lot of attention. However, the task is far from being solved: state-of-the-art performance on keyphrase extraction is still much lower than that on many core NLP tasks (Liu et al., 2010). Our goal in this paper is to survey the state of the art in keyphrase extraction, examining the major sources of errors made by existing systems and discussing the challenges ahead. 2 Corpora Automatic keyphrase extraction systems have been evaluated on corpora from a variety of sources ranging from long scientific publications to short paper abstracts and email messages. Ta- ble 1 presents a listing of the corpora grouped by their sources as well as their statistics.1 There are at least four corpus-related factors that affect the difficulty of keyphrase extraction. Length The difficulty of the task increases with the length of the input document as longer doc- uments yield more candidate keyphrases (i.e., phrases that are eligible to be keyphrases (see Sec- tion 3.1)). For instance, each Inspec abstract has on average 10 annotator-assigned keyphrases and 34 candidate keyphrases. In contrast, a scientific paper typically has at least 10 keyphrases and hun- dreds of candidate keyphrases, yielding a much bigger search space (Hasan and Ng, 2010). Conse- quently, it is harder to extract keyphrases from sci- entific papers, technical reports, and meeting tran- scripts than abstracts, emails, and news articles. Structural consistency In a structured doc- ument, there are certain locations where a keyphrase is most likely to appear. For instance, most of a scientific paper’s keyphrases should …

Content truncated to 500 words