LEHAN: Link-Feature Enhanced Heterogeneous Graph Attention Network

Graph Neural Networks (GNNs) have been studied extensively and have performed well in solving complex machine learning tasks in recent years. Many GNN-based approaches focused on representing homogeneous graphs with only a single type of nodes and links. However, many real-world networks are heterogeneous, involving various types of nodes and links. Existing GNN-based approaches for representing heterogeneous graphs only focused on node features and meta-paths, which often causes difficulties in reflecting link features to learn the graph representations. To overcome this limitation, we propose a Link-feature Enhanced Heterogeneous graph Attention Network (LEHAN) that focuses on the node and link features to represent heterogeneous graphs. LEHAN consists of the node attention block and the link attention block, where each block aggregates node features and link features by attention mechanism with meta-paths information. The extensive experimental evaluations show that LEHAN outperforms the state-of-the-art graph embedding algorithms in node classification and clustering on real-world heterogeneous graphs.

Automated summary

This article proposes a Link-feature Enhanced Heterogeneous graph Attention Network (LEHAN) for representing heterogeneous graphs in the real world. Experimental results show that LEHAN outperforms existing graph embedding algorithms in node classification and clustering.