Clustering to Find Exemplar Terms for Keyphrase Extraction

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 257–266, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Clustering to Find Exemplar Terms for Keyphrase Extraction Zhiyuan Liu, Peng Li, Yabin Zheng, Maosong Sun Department of Computer Science and Technology State Key Lab on Intelligent Technology and Systems National Lab for Information Science and Technology Tsinghua University, Beijing 100084, China {lzy.thu, pengli09, yabin.zheng}@gmail.com, sms@tsinghua.edu.cn Abstract Keyphrases are widely used as a brief summary of documents. Since man- ual assignment is time-consuming, vari- ous unsupervised ranking methods based on importance scores are proposed for keyphrase extraction. In practice, the keyphrases of a document should not only be statistically important in the docu- ment, but also have a good coverage of the document. Based on this observa- tion, we propose an unsupervised method for keyphrase extraction. Firstly, the method finds exemplar terms by leverag- ing clustering techniques, which guaran- tees the document to be semantically cov- ered by these exemplar terms. Then the keyphrases are extracted from the doc- ument using the exemplar terms. Our method outperforms sate-of-the-art graph- based ranking methods (TextRank) by 9.5% in F1-measure. 1 Introduction With the development of Internet, information on the web is emerging exponentially. How to effec- tively seek and manage information becomes an important research issue. Keyphrases, as a brief summary of a document, provide a solution to help organize, manage and retrieve documents, and are widely used in digital libraries and information re- trieval. Keyphrases in articles of journals and books are usually assigned by authors. However, most articles on the web usually do not have human-assigned keyphrases. Therefore, automatic keyphrase extraction is an important research task. Existing methods can be divided into supervised and unsupervised approaches. The supervised approach (Turney, 1999) re- gards keyphrase extraction as a classification task. In this approach, a model is trained to determine whether a candidate term of the document is a keyphrase, based on statistical and linguistic fea- tures. For the supervised keyphrase extraction approach, a document set with human-assigned keyphrases is required as training set. However, human labelling is time-consuming. Therefore, in this study we focus on unsupervised approach. As an example of an unsupervised keyphrase extraction approach, the graph-based ranking (Mi- halcea and Tarau, 2004) regards keyphrase extrac- tion as a ranking task, where a document is repre- sented by a term graph based on term relatedness, and then a graph-based ranking algorithm is used to assign importance scores to each term. Existing methods usually use term cooccurrences within a specified window size in the given document as an approximation of term relatedness (Mihalcea and Tarau, 2004). As we know, none of these existing works gives an explicit definition on what are appropri- ate keyphrases for a document. In fact, the existing methods only judge the importance of each term, and extract the most important ones as keyphrases. From the observation of human-assigned keyphrases, we conclude that good keyphrases of a document should satisfy the following properties: 1. Understandable. The …

Content truncated to 500 words