Str-GCL: Structural Commonsense Driven Graph Contrastive Learning

Dongxiao He, Yongqi Huang, Jitao Zhao, Xiaobao Wang, Zhen Wang

Track: Graph algorithms and modeling for the Web

Keywords: Graph Neural Networks, Self-Supervised Learning on Graphs, Graph Contrastive Learning, Structural Commonsense

Abstract: Graph Contrastive Learning (GCL) is a widely adopted approach in unsupervised representation learning, utilizing representational constraints to derive effective embeddings. However, current GCL methods primarily focus on capturing implicit semantic relationships, often overlooking the structural commonsense embedded within the graph’s structure and attributes. This structural commonsense is crucial for effective representation learning. Identifying and integrating such structural commonsense in GCL poses a significant challenge. To address this gap, we propose a novel framework called Structural Commonsense Unveiling in Graph Contrastive Learning (Str-GCL). Str-GCL leverages first-order symbolic logic rules to represent structural commonsense and explicitly integrates these rules into the GCL framework. Specifically, we introduce structural commonsense from both topological and attribute rule perspectives, processing these rules independently without modifying the original graph. Additionally, we design a representation alignment mechanism that guides the encoder to effectively capture this structural commonsense. To the best of our knowledge, this is the first attempt to directly incorporate structural commonsense into GCL in a rule-based manner. Extensive experiments demonstrate that Str-GCL significantly outperforms existing GCL methods, providing a new perspective on leveraging structural commonsense in graph representation learning.

Submission Number: 656