A Concentric-Circle Model for Community Mining

Discovering communities from a graph structure such as the Web has become an interesting research problem recently. In this paper, comparing with the state-of-the-art authority detecting and graph partitioning methods, we propose a concentric-circle model to more accurately define communities. With this model, a community could be described as a set of concentric-circles. The most important objects representing the concept of a whole community lie in the center and are called core objects. Affiliated objects, which are related to the core objects, surround the core with different ranks. Base on the concentric-circle model, a novel algorithm is developed to discover communities conforming to this model. We also conducted a case study to automatically discover research interest groups in the computer science domain from the Web. Experiments show that our method is very effective to generate high-quality communities with more clear structure and more tunable granularity.