MGRL: attributed multiplex heterogeneous network representation learning based on multi-granularity information fusion

Nowadays, attributed multiplex heterogeneous network (AMHN) representation learning has shown superiority in many network analysis tasks due to its ability to preserve both the structure of the network and the semantics of the nodes. However, few people consider the correlation between content attributes within each node. No personalized analysis model is designed for different semantics of heterogeneous relations. To address these issues, this paper proposes an MGRL model. MGRL adopts a filter based on variance discrimination to filter out the noise information in the node content attributes. To better utilize semantic characteristics of heterogeneous relations, personalized fusion models are designed according to heterogeneous relation categories: peer relations and subordinate relations. Results of experiments conducted on three real-world datasets show an obvious advantage of the proposed MGRL model over state-of-the-art baseline methods.