Multi-view heterogeneous graph representation learning with fusion of high-order information and low-order information

Graph neural networks (GNNs) are extensively employed in the analysis of data structured as graphs and excel at learning intricate node relationships. However, traditional GNN methods encounter difficulties when applied to analyzing heterogeneous graphs (HGs) with diverse node types and relationships, due to the complex heterogeneity and semantics of the data. Therefore, we propose Multi-view Heterogeneous Graph Representation Learning with Fusion of High-Order Information and Low-Order Information(MHGRL). MHGRL produces node embeddings by integrating node content transformation, meta-path-based intra- and inter-aggregation, as well as the amalgamation of low- and high-order information from multiple perspectives. Comprehensive experiments on three diverse heterogeneous graph datasets reveal the significant benefits of our MHGRL in node classification endeavors.