Learning Weight Signed Network Embedding with Graph Neural Networks

Abstract Network embedding aims to map nodes in a network low-dimensional vector representations. Graph neural networks (GNNs) have received much attention and achieved state-of-the-art performance learning node representation. Using fundamental sociological theories (status theory balance theory) model signed networks, basing GNN on has become hot topic embedding. However, most GNNs fail use edge weight information models cannot be directly used weighted networks. We propose novel directed graph named WSNN learn for Weighted The proposed reconstructs link signs, directions, triangles simultaneously. Based the representations learned by model, we conduct sign prediction Extensive experimental results real-world datasets demonstrate superiority of over algorithms representation