Multi-topology contrastive graph representation learning

Self-supervised graph representation learning has received significant attention by virtue of tackling the label scarcity issue in graph data. However, prior methods underutilize graph structures with multiple forms and subgraph structures at different scales, thus failing to deeply explore the diversity and complexity of graph data. In this paper, we present a novel multi-topology contrastive graph representation learning (MCGRL) framework, which aims to improve the effectiveness of node representation learning by capturing multi-granularity information in different topologies. Specifically, we generate multiple topologies from different viewpoints and then contrast the learned multi-granularity node representations in different topologies to preserve the rich multi-topology interactions and complementary information. Drawing upon an in-depth scrutiny of the classical Intersection over Union, we propose a subgraph-level similarity constraint (SIoU) to explore the semantic consistency among multiple topologies and dynamically characterize different-granularity subgraph information. Empirical experiments on real-world datasets demonstrate the effectiveness of our proposed method compared with current state-of-the-art methods.