Detecting Overlapping Community Structures in Networks with Global Partition and Local Expansion

The problem of discovering community structures in a network has received a lot of attention in many fields like social network, weblog, and protein-protein interaction network. Most of the efforts, however, were made to measure, qualify, detect, and refine "uncrossed" communities from a network, where each member in a network was implicitly assumed to play an unique role corresponding to its resided community. In practical, this hypothesis is not always reasonable. In social network, for example, one people can perform different interests and thus become members of multiple real communities. In this context, we propose a novel algorithm for finding overlapping community structures from a network. This algorithm can be divided into two phases: 1) globally collect proper seeds from which the communities are derived in next step; 2) randomly walk over the network from the seeds by a well designed local optimization process. We conduct the experiments by real-world networks. The experimental results demonstrate high quality of our algorithm and validate the usefulness of discovering overlapping community structures in a networks.