MFHE: Multi-View Fusion-Based Heterogeneous Information Network Embedding

Depending on the type of information network, network embedding is classified into homogeneous and heterogeneous (HIN) embedding. Compared with HIN composition more complex contains richer semantics. At present, research relatively mature. However, if model directly applied to HIN, it will cause incomplete extraction. It necessary build a specialized for HIN. Learning based meta-path an effective approach extracting semantic information. Nevertheless, only from single view loss. To solve these problems, we propose multi-view fusion-based model, called MFHE. MFHE includes four parts: node feature space transformation, subview extraction, fusion, training. divides different subviews meta-paths, models local accurately in multi-head attention mechanism, then fuses through spatial matrix. In this paper, consider relationship between subviews; thus, applicable Experiments are conducted ACM DBLP datasets. baselines, experimental results demonstrate that effectiveness has been improved.