Block Modeling-Guided Graph Convolutional Neural Networks

Graph Convolutional Network (GCN) has shown remarkable potential of exploring graph representation. However, the GCN aggregating mechanism fails to generalize networks with heterophily where most nodes have neighbors from different classes, which commonly exists in real-world networks. In order make propagation and aggregation suitable for both homophily (or even their mixture), we introduce block modelling into framework so that it can realize “block-guided classified aggregation”, automatically learn corresponding rules classes. By incorporating process, is able aggregate information homophilic heterophilic discriminately according degree. We compared our algorithm state-of-art methods deal problem. Empirical results demonstrate superiority new approach over existing datasets while maintaining a competitive performance datasets.