RW-HeCo: A random walk and network centrality based graph neural network for community detection in heterogeneous networks

Real-world networks often consist of different types of nodes, which leads to the creation of heterogeneous graphs. Most studies on heterogeneous graph neural networks follow the semi-supervised learning paradigm. The purpose of community detection in heterogeneous networks is to identify groups or communities of nodes that share similar characteristics or functions. In state-of-the-art community detection work, all meta-path-based neighbors were considered, but not all connections among meta-paths are necessary. In this paper, we propose a novel approach called RW-HeCo i.e. Random Walk and Network Centrality based GNN (Graph Neural Network) for Community Detection in Heterogeneous Networks. This approach uses a random walk and network centrality-based GNN along with co-contrastive learning. Our method is able to capture and categorize the structures more effectively and efficiently by adopting a network schema view and a meta path-based random walk. In our experiments, we evaluate the performance of RW-HeCo on four benchmark networks (ACM, AMiner, DBLP, and Freebase) and demonstrate improved classification accuracy that outperforms state-of-the-art methods. Moreover, to the best of our knowledge, the results obtained for ACM, DBLP, and Freebase datasets are the best compared to all the existing NMI (Normalized Mutual Information) and ARI (Adjusted Rand Index) values.