JGCL: Joint Self-Supervised and Supervised Graph Contrastive Learning

Abstract

Semi-supervised and self-supervised learning on graphs are two popular avenues for graph representation learning. We demonstrate that no single method from semi-supervised and self-supervised learning works uniformly well for all settings in the node classification task. Self-supervised methods generally work well with very limited training data, but their performance could be further improved using the limited label information. We propose a joint self-supervised and supervised graph contrastive learning (JGCL) to capture the mutual benefits of both learning strategies. JGCL utilizes both supervised and self-supervised data augmentation and a joint contrastive loss function. Our experiments demonstrate that JGCL and its variants are one of the best performers across various proportions of labeled data when compared with state-of-the-art self-supervised, unsupervised, and semi-supervised methods on various benchmark graphs.