SimGRL: a simple self-supervised graph representation learning framework via triplets

Abstract Recently, graph contrastive learning (GCL) has achieved remarkable performance in representation learning. However, existing GCL methods usually follow a dual-channel encoder network (i.e., Siamese networks), which adds to the complexity of architecture. Additionally, these overly depend on varied data augmentation techniques, corrupting information. Furthermore, they are heavily reliant large quantities negative nodes for each object node, requires tremendous memory costs. To address issues, we propose novel and simple framework, named SimGRL. Firstly, our proposed architecture only contains one based neural instead encoder, simplifies Then introduce distributor generate triplets obtain views between their neighbors, avoiding need augmentations. Finally, design triplet loss adjacency information graphs that utilizes node reducing overhead significantly. Extensive experiments demonstrate SimGRL achieves competitive classification tasks, especially terms running time overhead.