GRNN: Graph-Retraining Neural Network for Semi-Supervised Node Classification

In recent years, graph neural networks (GNNs) have played an important role in representation learning and successfully achieved excellent results semi-supervised classification. However, these GNNs often neglect the global smoothing of because is incompatible with node Specifically, a cluster nodes has small number other classes nodes. To address this issue, we propose graph-retraining network (GRNN) model that performs over by alternating between procedure inference procedure, based on key idea expectation-maximum algorithm. Moreover, error combined cross-entropy to form loss function GRNN, which effectively solves problem. The experiments show GRNN achieves high accuracy standard citation datasets, including Cora, Citeseer, PubMed, proves effectiveness