Encoder augmentation for multi-task graph contrastive learning

Xiaoyu Wang, Qiqi Zhang, Gen Liu, Zhongying Zhao, Hongzhi Cui

Abstract: Graph contrastive learning (GCL), as one of the most popular self-supervised paradigms, has achieved a great deal of success in the field of graph representation learning. However, many GCL methods face a limitation in generating diverse enough views due to the absence of prior knowledge, leading to suboptimal performance. Moreover, the community structures are rarely considered. We argue that the semantic similarities among nodes within the same community substantially contribute to graph analyzing tasks. To address these issues, we propose an Encoder Augmentation for Multi-task Graph Contrastive Learning method, known as EA-MGCL. Specifically, we introduce a multi-layer information weighting aggregation method to enhance the diversity of the generated views within the encoder augmentation module. Additionally, we present a multi-task graph contrastive learning module to discern local community information. The proposed EA-MGCL effectively captures and leverages node-level, node-community-level, and community-level information via three contrastive learning strategies. Extensive experimental results on six real-world datasets demonstrate that EA-MGCL achieves superior performance compared to state-of-the-art methods. The codes of this work are available at https://github.com/ZZY-GraphMiningLab/EA-MGCL.