Combining Graph Convolutional Neural Networks and Label Propagation

Label Propagation Algorithm (LPA) and Graph Convolutional Neural Networks (GCN) are both message passing algorithms on graphs. Both solve the task of node classification, but LPA propagates label information across edges graph, while GCN transforms feature information. However, conceptually similar, theoretical relationship between has not yet been systematically investigated. Moreover, it is unclear how can be combined under a unified framework to improve performance. Here we study in terms feature/label influence , which characterize much initial one influences final another GCN/LPA. Based our analysis, propose an end-to-end model that combines LPA. In model, edge weights learnable, serves as regularization assist learning proper lead improved Our also seen based labels, more direct efficient than existing feature-based attention models or topology-based diffusion models. number experiments for semi-supervised classification knowledge-graph-aware recommendation, shows superiority over state-of-the-art baselines.