Abstract: Graph contrastive learning (GCL) leverages semantic consistency as contrastive signals and has shown strong performance in semi-supervised node classification. However, real-world graphs ...