Stochastic block models for community detection in heterogeneous networks

Heterogeneous networks have multiple types of nodes and edges. Community detection in single layer and multiplex networks has been extensively studied in the past decade. But there are few methods have constructed for heterogeneous networks. In this paper, we introduce heterogeneous stochastic block models for detecting communities in heterogeneous networks. Generally these models are developed based on generalization of single-layer stochastic block model, bipartite stochastic block model and multiplex stochastic block model. We define this two types of stochastic block model, independent degree and shared degree. Independent degree models have one specific degree parameter for each layers, shared degree models share one degree parameter for all layers. We introduce a method to create synthetic networks with benchmark heterogeneous communities. We evaluate the performance of the proposed community detection algorithm with generalization of Kernighan-Lin algorithm in the controlled environment (with synthetic benchmark communities). According to our results, shared degree models have better performance in high crossed networks in contrast independent degree models have better performance in low crossed networks. Exception when intra-layer densities are high and inter-layer densities are low, single-layer algorithm (flattering network) has better performance. On real datasets, DBLP and AMiner four-area datasets, proposed methods have good results.