Signed Laplacian Graph Neural Networks

This paper studies learning meaningful node representations for signed graphs, where both positive and negative links exist. problem has been widely studied by meticulously designing expressive graph neural networks, as well capturing the structural information of through traditional structure decomposition methods, e.g., spectral theory. In this paper, we propose a novel representation framework, called Signed Laplacian Graph Neural Network (SLGNN), which combines advantages both. Specifically, based on theory signal processing, first design different low-pass high-pass convolution filters to extract low-frequency high-frequency links, respectively, then combine them into unified message passing framework. To effectively model further self-gating mechanism estimate impacts during passing. We mathematically establish relationship between aggregation process in SLGNN regularization theoretically analyze expressiveness SLGNN. Experimental results demonstrate that outperforms various competitive baselines achieves state-of-the-art performance.