Discovering Densest Subgraph over Heterogeneous Information Networks

Densest Subgraph Discovery (DSD) is a fundamental and challenging problem in the field of graph mining in recent years. The DSD aims to determine, given a graph G, the subgraph with the maximum density according to the specified definition. The majority of current research focuses on identifying DSD in homogeneous information networks. Nevertheless, these techniques cannot be applied directly in Heterogeneous Information Networks (HINs) since the semantics of paths are not considered. This limits the application of these approaches in certain fields, as many graphs, e.g., the DBLP dataset, comprise of many types of edges. In order to remedy this research need, this paper proposes approaches for resolving the DSD issue over HINs. By examining numerous linkage paths between two vertices, we characterize the relationship between two objects of the same type using the concept of meta-path. We further build two new HIN models and provide DSD techniques that perform well in large networks. In addition, comprehensive experiments on various HIN datasets demonstrate the effectiveness and efficiency of the proposed methods.