Path-Graph Fusion Based Community Detection over Heterogeneous Information Network

Abstract

As a natural and general representation of data in the real world, heterogeneous information network (HIN) has been employed to model complex and heterogeneous data in many tasks. Community detection on heterogeneous network has received much attention in recent years. Most of HIN based methods rely on meta-path based similarity. But there are two challenges exist in the approaches. One is the similarity measure directly obtained by a meta-path is often a bias measure. The other is how to effectively aggregate different meta-path based similarities for clustering. In this paper, we propose a path-graph fusion based community detection model called PGFCluster. Our model utilizes a PathSim-based normalization to eliminate similarity bias. In addition, we design a flexible fusion mechanism with dynamically optimizing fusion result for best community partition. Experiments on two real-world datasets demonstrate the effectiveness of our model compared to other methods.