SDGNN: Symmetry-Preserving Dual-Stream Graph Neural Networks

Abstract

Dear Editor, This letter proposes a symmetry-preserving dual-stream graph neural network (SDGNN) for precise representation learning to an undirected weighted graph (UWG). Although existing graph neural networks (GNNs) are influential instruments for representation learning to a UWG, they invariably adopt a unique node feature matrix for illustrating the sole node set of a UWG. Such a modeling strategy can limit the representation learning ability due to the diminished feature space. To this end, the proposed SDGNN innovatively adopts the following two-fold ideas: 1) Building a dual-stream graph learning framework that tolerates multiple node feature matrices for boosting the representation learning ability; 2) Integrating a symmetry regularization term into the learning objective for implying the equality constraint among its multiple node feature matrices, which exemplifies a graph's intrinsic symmetry and prompts learning the multiple node embeddings jointly. Experiments on six real-world UWG datasets indicate that the proposed SDGNN has superior performance in addressing the task of missing link estimation compared with the state-of-the-art baselines.