TITPI: Web People Search Task Using Semi-Supervised Clustering Approach

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 318–321, Prague, June 2007. c⃝2007 Association for Computational Linguistics TITPI: Web People Search Task Using Semi-Supervised Clustering Approach Kazunari Sugiyama Precision and Intelligence Laboratory Tokyo Institute of Technology 4259 Nagatsuta, Midori, Yokohama, Kanagawa 226-8503, Japan sugiyama@lr.pi.titech.ac.jp Manabu Okumura Precision and Intelligence Laboratory Tokyo Institute of Technology 4259 Nagatsuta, Midori, Yokohama, Kanagawa 226-8503, Japan oku@pi.titech.ac.jp Abstract Most of the previous works that disam- biguate personal names in Web search re- sults employ agglomerative clustering ap- proaches. However, these approaches tend to generate clusters that contain a single el- ement depending on a certain criterion of merging similar clusters. In contrast to such previous works, we have adopted a semi- supervised clustering approach to integrate similar documents into a labeled document. Moreover, our proposed approach is char- acterized by controlling the fluctuation of the centroid of a cluster in order to generate more accurate clusters. 1 Introduction Personal names are often submitted to search en- gines as query keywords, as described in a report1 indicating that about 10% of the English queries from the search engine ALLTheWeb2 contain per- sonal names. However, in response to a personal name query, search engines return a long list of search results containing that contains Web pages about several namesakes. For example, when a user submits a personal name like “William Cohen” as a query to the search engine Google3, the re- turned results represent more than one person named “William Cohen.” In the results, a computer sci- ence professor, an American politician, a surgeon, 1http://tap.stanford.edu/PeopleSearch.pdf 2http://www.alltheweb.com/ 3http://www.google.com/ and others are not classified into separate clusters but mixed together. Most of the previous works on disambiguating personal names in Web search results employ sev- eral kinds of agglomerative clustering approach as described in Section 2. However, in these ap- proaches, a lot of clusters that contain only one el- ement tend to be generated, depending on a certain criterion for merging similar clusters. In addition, in person search results from the World Wide Web (WWW), we can often observe that a small num- ber of entities have a lot of search-result Web pages, while others have only one or two. In light of these facts, if a labeled Web page that describes a person is introduced, clustering for personal name disam- biguation would be much more accurate. In the fol- lowing, we refer to such a labeled Web page as the “seed page.” Then, in order to disambiguate per- sonal names in Web search results, we introduce semi-supervised clustering that uses the seed page to aid the clustering of unlabeled search-result Web pages. Our semi-supervised clustering approach is characterized by controlling the fluctuation of the 2centroid of a cluster. Related Work (Mann and Yarowsky, 2003) first extract biographi- cal information, such as birthdates, birthplaces, oc- cupations, and so on. Then, for each document, they generate a feature vector composed of the ex- tracted biographical information, proper nouns, and the TF-IDF score computed from the documents …

Content truncated to 500 words