Fuzzy Ants Clustering for Web People Search
Els Lefever, Timur Fayruzov, Véronique Hoste, Martine De Cock · Ghent University Academic Bibliography (Ghent University) · 2009
A search engine query for a person’s name often brings up web pages corresponding to several people who share the same name. The Web People Search (WePS) problem involves organizing such search results for an ambiguous name query in meaningful clusters, that group together all web pages corresponding to one single individual. A particularly challenging aspect of this task is that it is in general not known beforehand how many clusters to expect. In this paper we therefore propose the use of a Fuzzy Ants clustering algorithm that does not rely on prior knowledge of the number of clusters that need to be found in the data. An evaluation on benchmark data sets from SemEval’s WePS1 and WePS2 competitions shows that the resulting system is competitive with the agglomerative clustering Agnes algorithm. This is particularly interesting as the latter involves manual setting of a similarity threshold (or estimating the number of clusters in advance) while the former does not.