|
Abstract:
|
Community detection for large networks is a challenging task due to the large community number as well as the computational cost. Among communities, the disassortative communities structure is hard to detect because the nodes across disassortative communities are densely connected. In this paper, we denote group structure as a group of disassortative communities while the group level remains as assortative. We incorporate the group structure to the stochastic blockmodel (SBM) and propose a novel divide-and-conquer algorithm to detect the community structure. The theoretical results have two parts, the first part is that the proposed method can recover the group structure asymptotically, the second part is the consistency of the estimated community label under certain regularity conditions. Moreover, numerical studies demonstrate that the proposed method can reduce the computational cost significantly while still achieving competitive performance for community detection.
|
