UNN-WePS: Web Person Search using co-Present Names and Lexical Chains

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 402–405, Prague, June 2007. c⃝2007 Association for Computational Linguistics UNN-WePS: Web Person Search using co-Present Names and Lexical Chains Jeremy Ellman Northumbria University Pandon Building Newcastle upon Tyne UK Jeremy. Ellman @unn.ac.uk Gary Emery Northumbria University Pandon Building Newcastle upon Tyne UK Gary.Emery@unn.ac.uk Abstract We describe a system, UNN-WePS for identifying individuals from web pages us- ing data from Semeval Task 13. Our sys- tem is based on using co-presence of per- son names to form seed clusters. These are then extended with pages that are deemed conceptually similar based on a lexical chaining analysis computed using Roget’s thesaurus. Finally, a single link hierarchical agglomerative clustering algorithm merges the enhanced clusters for individual entity recognition. UNN-WePS achieved an aver- age purity of 0.6, and inverse purity of 0.73. 1 Introduction Guha and Garg (2004) report that approximately 4% of internet searches are to locate named indi- viduals. Yet, many people share the same name with for example 157630 individuals in the UK sharing the most common name ‘David Jones’ (UK statistics cited by Ash 2006). Consequently identifying web pages on specific individuals is a significant problem that will grow as everyone ac- quires a web presence. There are several proposed approaches to identi- fying which individuals correspond to which web pages. For example, Bollegala et al. (2007) pro- pose augmenting queries in the style of relevance feedback (Salton and Buckley 1990), Kalashnikov (2007) treat Web Person Search (WePS) as a dis- ambiguation problem whose objective is to distin- guish individuals, whilst Wan et al. (2005) see WePS as a clustering problem. WePS has both similarities and differences to word sense disambiguation (WSD). Both seek to classify instances of usage, but in WSD the sense inventory is fixed. WSD then is more amenable to a classification solution where a system can be ef- fectively trained using learning algorithms. In WePS we do not know from the outset how many individuals our pages correspond to. Consequently we took the view that WePS is better seen as a clustering rather than a classification problem. 1.1 Ambiguity Ambiguity is a common feature of WePS and WSD. There are multiple types of ambiguity in the relation between person names and entities that confound overly simple approaches. Firstly, note that some first names are also last names (Les Paul, Michael Howard), and that some last names also occur as given names (Woodrow Wilson Guthrie, Martin Luther King). Consequently, an overly simple name parser will easily be confused. Sec- ondly many last names are also place names (Jack London, William Manchester). Thus, if a last name is not found in the names database, but is found in the gazetteer, a name can be confused with a loca- tion. Finally, we come to toponym ambiguity, where the name of a place may correspond to sev- eral locations. (For example, there are thirteen places called Manchester, multiple Londons, Washingtons etc.) Resolving toponyms is a re- search problem itself …

Content truncated to 500 words