Experiments on Semantic-based Clustering for Cross-document Coreference

Experiments on Semantic-based Clustering for Cross-document Coreference Horacio Saggion Department of Computer Science University of Sheffield 211 Portobello Street - Sheffield, England, UK, S1 4DP Tel: +44-114-222-1947 Fax: +44-114-222-1810 saggion@dcs.shef.ac.uk Abstract We describe clustering experiments for cross-document coreference for the first Web People Search Evaluation. In our ex- periments we apply agglomerative cluster- ing to group together documents potentially referring to the same individual. The algo- rithm is informed by the results of two dif- ferent summarization strategies and an off- the-shelf named entity recognition compo- nent. We present different configurations of the system and show the potential of the ap- plied techniques. We also present an analy- sis of the impact that semantic information and text summarization have in the cluster- ing process. 1 Introduction Finding information about people on huge text col- lections or on-line repositories on the Web is a com- mon activity. In ad-hoc Internet retrieval, a request for documents/pages referring to a person name may return thousand of pages which although containing the name, do not refer to the same individual. Cross- document coreference is the task of deciding if two entity mentions in two sources refer to the same indi- vidual. Because person names are highly ambiguous (i.e., names are shared by many individuals), decid- ing if two documents returned by a search engine such as Google or Yahoo! refer to the same individ- ual is a difficult problem. Automatic techniques for solving this problem are required not only for better access to information but also in natural language processing applications such as multidocument summarization, question an- swering, and information extraction. Here, we con- centrate on the Web People Search Task (Artiles et al., 2007) as defined in the SemEval 2007 Work- shop: a search engine user types in a person name as a query. Instead of ranking web pages, an ideal sys- tem should organise search results in as many clus- ters as there are different people sharing the same name in the documents returned by the search en- gine. The input is, therefore, the results given by a web search engine using a person name as query. The output is a number of sets, each containing doc- uments referring to the same individual. The task is related to the coreference resolution problem disre- garding however the linking of mentions of the tar- get entity inside each single document. Similarly to (Bagga and Baldwin, 1998; Phan et al., 2006), we have addressed the task as a document clustering problem. We have implemented our own clustering algorithms but rely on available extraction and summarization technology to produce document representations used as input for the clustering pro- cedure. We will shown that our techniques produce not only very good results but are also very compet- itive when compared with SemEval 2007 systems. We will also show that carefully selection of docu- ment representation is of paramount importance to achieve good performance. Our system has a sim- ilar level of performance as the best system in …

Content truncated to 500 words