Modularity for Bipartite Networks.

Several real-world data are represented as bipartite networks composed of two types of vertices, such as paper–author networks and event–attendee networks. Discovering communities from such bipartite networks is important for finding similar items and for understanding overall network structures. In order to evaluate the quality of divisions of normal (unipartite) networks, Newman’s modularity is widely used. Recently, modularities for bipartite networks are proposed by Guimera and Barber. These bipartite modularities are, however, not sufficient for evaluating the degree of correspondence between communities of different vertex types, which is often important for understanding the characteristics of the communities. For example, close-knit paper communities and author communities indicate that their research topics are relatively focused rather than loose-knit communities. This chapter proposes a new bipartite modularity for evaluating community extraction from bipartite networks. Experimental results show that our new bipartite modularity is appropriate for discovering close-knit communities, and it is also useful for characterizing the communities.