Multiple heterogeneous network representation learning based on multi-granularity fusion

Heterogeneous network representation learning shows its superior capacity in complex network analysis. It aims to embed nodes into a low-dimensional space and pursues a meaningful vector representation for each node. At present, the research of heterogeneous networks mainly focuses on the fusion of network structure information, semantic information, and attribute information. However, how to effectively extract neighboring structure information and accurately extract semantic information are still open problems. Furthermore, few people consider the difference between the importance of low-order structural information and high-order semantic information. In this paper, a multiple heterogeneous network representation learning framework based on multi-granularity information fusion called MHRL is proposed to solve these problems. MHRL considers the structural and semantic information of the heterogeneous network as different information grains, and uses different encoders to obtain structural and semantic embeddings. Then, MHRL uses a biased contrastive fusion method to effectively fuse structural embeddings and semantic embeddings. Extensive experiments on three real-world datasets show that the proposed method is significantly better than the state-of-the-art baselines in classification, clustering and visualization.