Eliciting Structural and Semantic Global Knowledge in Unsupervised Graph Contrastive Learning

Graph Contrastive Learning (GCL) has recently drawn much research interest for learning generalizable node representations in a self-supervised manner. In general, the contrastive process GCL is performed on top of learned by graph neural network (GNN) backbone, which transforms and propagates contextual information based its local neighborhoods. However, nodes sharing similar characteristics may not always be geographically close, poses great challenge unsupervised efforts due to their inherent limitations capturing such global knowledge. this work, we address proposing simple yet effective framework -- Simple Neural Networks with Structural Semantic Learning} (S^3-CL). Notably, virtue proposed structural semantic algorithms, even can learn expressive that preserve valuable patterns. Our experiments demonstrate S^3-CL) achieve superior performance different downstream tasks compared state-of-the-art methods. Implementation more experimental details are publicly available at https://github.com/kaize0409/S-3-CL.