An MDL Approach to Efficiently Discover Communities in Bipartite Network

Bipartite network is a branch of complex network. It is widely used in many applications such as social network analysis, collaborative filtering and information retrieval. Partitioning a bipartite network into smaller modules helps to get insight of the structure of the bipartite network. The main contributions of this paper include: (1) proposing an MDL 21 criterion for identifying a good partition of a bipartite network. (2) presenting a greedy algorithm based on combination theory, named as MDL-greedy, to approach the optimal partition of a bipartite network. The greedy algorithm automatically searches for the number of partitions, and requires no user intervention. (3) conducting experiments on synthetic datasets and the southern women dataset. The results show that our method generates higher quality results than the state-of-art methods Cross-Association and Information-theoretic co-clustering. Experiment results also show the good scalability of the proposed algorithm. The highest improvement could be up to about 14% for the precision, 40% for the ratio and 70% for the running time.