Concept hierarchy construction by combining spectral clustering and subsumption estimation

Abstract

With the rapid development of the Web, how to add structural guidance (in the form of concept hierarchies) for Web document navigation becomes a hot research topic. In this paper, we present a method for the automatic acquisition of concept hierarchies. Given a set of concepts, each concept is regarded as a vertex in an undirected, weighted graph. The problem of concept hierarchy construction is then transformed into a modified graph partitioning problem and solved by spectral methods. As the undirected graph cannot accurately depict the hyponymy information regarding the concepts, subsumption estimation is introduced to guide the spectral clustering algorithm. Experiments on real data show very encouraging results. © Springer-Verlag Berlin Heidelberg 2006.