Topical analysis for identification of web communities

Abstract

Traditional link-based schemes for identification of web communities focus on partitioning the web graph more sophisticatedly, without concerning the topical information inherently held by web pages. In this paper, we give a novel method of measuring the topicality of a hyperlink according to its context. Based on this, we propose a topical maxflow-mincut algorithm which incorporates topical information into the traditional maxflow-mincut algorithm. Experiments show that our algorithm outperforms the traditional algorithm in identifying high-quality web communities. © Springer-Verlag Berlin Heidelberg 2009.