Multiplex Heterogeneous Graph Convolutional Network

Abstract

Heterogeneous graph convolutional networks have gained great popularity in\ntackling various network analytical tasks on heterogeneous network data,\nranging from link prediction to node classification. However, most existing\nworks ignore the relation heterogeneity with multiplex network between\nmulti-typed nodes and different importance of relations in meta-paths for node\nembedding, which can hardly capture the heterogeneous structure signals across\ndifferent relations. To tackle this challenge, this work proposes a Multiplex\nHeterogeneous Graph Convolutional Network (MHGCN) for heterogeneous network\nembedding. Our MHGCN can automatically learn the useful heterogeneous meta-path\ninteractions of different lengths in multiplex heterogeneous networks through\nmulti-layer convolution aggregation. Additionally, we effectively integrate\nboth multi-relation structural signals and attribute semantics into the learned\nnode embeddings with both unsupervised and semi-supervised learning paradigms.\nExtensive experiments on five real-world datasets with various network\nanalytical tasks demonstrate the significant superiority of MHGCN against\nstate-of-the-art embedding baselines in terms of all evaluation metrics.\n