Graph Contrastive Learning with Cohesive Subgraph Awareness
Keywords: social networks, graph contrastive learning, self-supervised learning, cohesive subgraph
TL;DR: We propose a novel graph contrastive learning method by preserving cohesion properties of graphs during the topology augmentation and graph learning processes.
Abstract: Graph contrastive learning (GCL) has emerged as a state-of-the-art strategy for learning representations of diverse graphs including social and biomedical networks. GCL widely uses stochastic graph topology augmentation, such as uniform node removal, to generate augmented graphs. However, such stochastic augmentations may severely damage the intrinsic properties of a graph and deteriorate the following representation learning process. Specifically, cohesive topological properties (e.g., $k$-core and $k$-truss) indicate strong and critical connections among multiple nodes; randomly removing nodes from a cohesive subgraph may remarkably alter the graph properties. In contrast, we argue that incorporating an awareness of cohesive subgraphs during the graph augmentation and learning processes has the potential to enhance GCL performance. To this end, we propose a novel unified framework called \textit{CTAug}, to seamlessly integrate cohesion awareness into various existing GCL mechanisms. In particular, \textit{CTAug} comprises two specialized modules: \textit{topology augmentation enhancement} and \textit{graph learning enhancement}. The former module generates augmented graphs that carefully preserve cohesion properties, while the latter module bolsters the graph encoder's ability to discern subgraph patterns. Theoretical analysis shows that \textit{CTAug} can strictly improve existing GCL mechanisms. Empirical experiments verify that \textit{CTAug} can achieve state-of-the-art performance for both graph and node representation learning, especially for graphs with high degrees.
Track: Graph Algorithms and Learning for the Web
Submission Guidelines Scope: Yes
Submission Guidelines Blind: Yes
Submission Guidelines Format: Yes
Submission Guidelines Limit: Yes
Submission Guidelines Authorship: Yes
Student Author: Yes
Submission Number: 950
Loading