Using Markov models for web site link prediction

Abstract

Markov models have been extensively used to model Web users' navigation behaviors on Web sites. The link structure of a Web site can be seen as a citation network. By applying bibliographic co-citation and coupling analysis to a Markov model constructed from a Web log file on a Web site, we propose a clustering algorithm called CitationCluster to cluster conceptually related pages. The clustering results are used to construct a conceptual hierarchy of the Web site. Markov model based link prediction is integrated with the hierarchy to assist users' navigation on the Web site.