Subgraph Adaptive Structure-Aware Graph Contrastive Learning

Graph contrastive learning (GCL) has been subject to more attention and widely applied numerous graph tasks such as node classification link prediction. Although it achieved great success even performed better than supervised methods in some tasks, most of them depend on node-level comparison, while ignoring the rich semantic information contained topology, especially for social networks. However, a higher-level comparison requires subgraph construction encoding, which remain unsolved. To address this problem, we propose adaptive structure-aware method (PASCAL) work, is subgraph-level GCL method. In PASCAL, construct subgraphs by merging all motifs that contain target node. Then encode basis motif number distribution capture hidden subgraphs. By incorporating information, PASCAL can richer local structures compared with other methods. Extensive experiments six benchmark datasets show outperforms state-of-art cases.