Data Augmentation for Graph Neural Networks

Data augmentation has been widely used to improve generalizability of machine learning models. However, comparatively little work studies data for graphs. This is largely due the complex, non-Euclidean structure graphs, which limits possible manipulation operations. Augmentation operations commonly in vision and language have no analogs Our graph neural networks (GNNs) context improving semi-supervised node-classification. We discuss practical theoretical motivations, considerations strategies augmentation. shows that edge predictors can effectively encode class-homophilic promote intra-class edges demote inter-class given structure, our main contribution introduces GAug framework, leverages these insights performance GNN-based node classification via prediction. Extensive experiments on multiple benchmarks show improves across GNN architectures datasets.