CollabRank: Towards a Collaborative Approach to Single-Document Keyphrase Extraction

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 969–976 Manchester, August 2008 CollabRank: Towards a Collaborative Approach to Single-Document Keyphrase Extraction Xiaojun Wan and Jianguo Xiao Institute of Computer Science and Technology Peking University, Beijing 100871, China {wanxiaojun,xiaojianguo}@icst.pku.edu.cn Abstract Previous methods usually conduct the keyphrase extraction task for single docu- ments separately without interactions for each document, under the assumption that the documents are considered inde- pendent of each other. This paper pro- poses a novel approach named Col- labRank to collaborative single-document keyphrase extraction by making use of mutual influences of multiple documents within a cluster context. CollabRank is implemented by first employing the clus- tering algorithm to obtain appropriate document clusters, and then using the graph-based ranking algorithm for col- laborative single-document keyphrase ex- traction within each cluster. Experimental results demonstrate the encouraging per- formance of the proposed approach. Dif- ferent clustering algorithms have been investigated and we find that the system performance relies positively on the qual- ity of document clusters. 1 Introduction A keyphrase is defined as a meaningful and sig- nificant expression consisting of one or more words in a document. Appropriate keyphrases can be considered as a highly condensed sum- mary for a document, and they can be used as a label for the document to supplement or replace the title or summary, thus facilitating users’ fast browsing and reading. Moreover, document key- phrases have been successfully used in the fol- lowing IR and NLP tasks: document indexing (Gutwin et al., 1999), document classification (Krulwich and Burkey, 1996), document cluster- © 2008. Licensed under the Creative Commons Attri- bution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc- sa/3.0/). Some rights reserved. ing (Zhang et al., 2004; Hammouda et al., 2005) and document summarization (Berger and Mittal, 2000; Buyukkokten et al., 2001). Keyphrases are usually manually assigned by authors, especially for journal or conference arti- cles. However, the vast majority of documents (e.g. news articles, magazine articles) do not have keyphrases, therefore it is beneficial to automatically extract a few keyphrases from a given document to deliver the main content of the document. Here, keyphrases are selected from within the body of the input document, without a predefined list (i.e. controlled vocabu- lary). Most previous work focuses on keyphrase extraction for journal or conference articles, while this paper focus on keyphrase extraction for news articles because news article is one of the most popular document genres on the web and most news articles have no author-assigned keyphrases. Very often, keyphrases of all single documents in a document set are required to be extracted. However, all previous methods extract key- phrases for a specified document based only on the information contained in that document, such as the phrase’s TFIDF, position and other syntac- tic information in the document. One common assumption of existing methods is that the docu- ments are independent of each other. Hence the keyphrase extraction task is conducted separately without interactions for each document. However, the multiple documents within an appropriate cluster …

Content truncated to 500 words