Detecting communities using bibliographic metrics

Abstract

We propose an efficient and novel approach for discovering communities in real-world random networks. Communities are formed by subsets of nodes in a graph, which are closely related. Extraction of these communities facilitates better understanding of such networks. Community related research has focused on two main problems: community discovery and community identification. Community discovery is the problem of extracting all the communities in a given network where as community identification is the problem of identifying the community to which a given set of nodes from the network belong. In this paper we first give a brief survey of the existing community-discovery algorithms and then propose a novel algorithm to discovering communities using bibliographic metrics. We also test the proposed algorithm on real-world networks and on computer-generated models with known community structures. Index Terms—Community discovery/identification, graph clustering. I. INTRODUCTION Recent studies on real-world random networks have revealed several interesting and significant properties like degree distribution, average distance between pairs of nodes, and network transitivity. These properties differentiate real- world networks from the classical Erdos and Renyi random graphs. One such property is community structure. These networks are locally dense but globally sparse. Each of these locally dense regions may be viewed a community. The term community has been defined in more than one way. Initially cliques and near-cliques were used to define communities with the idea that high connectivity corresponds to similarity between those nodes. Kleinberg while studying web graphs introduced the concept of and authorities. Authorities are web pages which are highly referenced and hubs are web pages that reference many authority pages. Later Gibson, Kleinberg and Raghavan define communities in web graph as a core of central, authoritative pages connected together by hub pages (1). Kumar, et al. define communities as bipartite cores: a bipartite core in a graph G consists of two (not necessarily disjoint) sets of nodes L and R, such that every node in L is adjacent to every node in