Missing nodes detection on graphs with self-supervised contrastive learning

Missing node detection in graphs is a problem of great significance in areas such as network mining and knowledge graph reasoning, as the graphs we obtain are often incomplete. Missing node detection requires identifying unobserved nodes and the edges between them and the observed nodes. Traditional approaches are limited by specific prior assumptions, preventing them from effectively leveraging the rich structural and attribute information in graphs or lacking an end -to -end framework. To address these challenges, we propose a novel method for identifying missing nodes in graphs using self -supervised contrastive learning, termed DMNG (Detection Missing Nodes in Graphs). Specifically, DMNG accomplishes missing node detection by two tasks of missing detection and node discrimination. The missing detection task is used to determine whether missing nodes exist between observed node pairs. The node discrimination task is used to reconstruct the edges between missing and observed nodes by merging of detected missing nodes, that is, determining whether different missing nodes are the same missing node. We designed two contrastive learning tasks to achieve these goals. DMNG trains these two tasks jointly with a multi -task loss in an end -to -end manner, allowing it to leverage information from both tasks to improve the performance on missing node detection. Extensive experiments on real -world benchmark datasets show the advantageous performance of DMNG over strong baseline methods.