Mining parameters that characterize the communities in web-like networks

Abstract

Community mining in large, complex, real-life networks such as the World Wide Web has emerged as a key data mining problem with important applications. In recent years, several graph theoretic definitions of community, generally motivated by empirical observations and intuitive arguments, have been put forward. However, a formal evaluation of the appropriateness of such definitions has been lacking. We present a new framework developed to address this issue, and then discuss a particular implementation of this framework. Finally, we present a set of experiments aimed at evaluating the effectiveness of two specific graph theoretic structures—alliance and near-clique—in capturing the essential properties of communities.