Community-Enhanced Contrastive Siamese Networks for Graph Representation Learning

Graph representation learning is the encoding of graph nodes into a low-dimensional representation space, which can effectively improve graph information representation while reducing the information dimensionality. To overcome the heavy reliance on label information of previous graph representation learning, the graph contrastive learning method has received attention from researchers as a self-supervised learning method, but it introduces the problem of sample selection dependence. To address this issue, inspired by deep clustering methods and image contrastive learning methods, we propose a novel Siamese network method, namely Community-enhanced Contrastive Siamese networks for Graph Representation Learning (MEDC). Specifically, we employ a Siamese network architecture to contrast two augmented views of the original graph and guide the network training by minimizing the similarity of positive sample nodes and negative sample nodes. Meanwhile, to take full advantage of the potential community structure of graph, we add a deep clustering layer in the network architecture, and the perceived community structure information is used to guide the selection of positive and negative samples. To demonstrate the effectiveness of the proposed method, we conducted a series of comparative experiments on three real datasets to validate the performance of our method.