Community Division of Heterogeneous Networks

Many real world data can be represented as heterogeneous networks that are composed of more than one types of nodes, such as paper-author networks (two types) and user-resource-tag networks (three types) of social tagging systems. Discovering communities from such heterogeneous networks is important for finding similar nodes, which are useful for information recommendation and visualization. Although modularity is a famous criterion for evaluating division of given networks, it is not applicable to heterogeneous networks. This paper proposes new modularity for bipartite networks, as the first step for heterogeneous networks. Experimental results using artificial networks and real networks are shown.