A divide-and-conquer approach to detecting latent community of practice from virtual organizations

Social network analysis methods have been exploited to support efficient collaborations in virtual organizations. However, a social network within a virtual organization is simply assumed to be homogeneous, i.e., all linkages between actors are contextually identical. For example, in bibliometrics, all linkages on a network are identical to "co-authoring" relationship between the actors. In this paper, we focus on integrating multiple social networks of which relationships between actors are heterogeneous. It makes a new relationship between two actors in different social networks possible to be discovered. In particular, we show how to detect latent community of practice from the multiple networks by measuring semantic centrality of actors. Thereby, we propose a divide-and-conquer approach based on the context matching algorithm, which is capable of separating the multiple social networks, with respect to the contexts of practice. We also take into account the relationships between topological features and the labels by statistical cooccurrence analysis.