Locally and Structurally Private Graph Neural Networks

Graph Neural Networks (GNNs) are known to address such tasks over graph-structured data, which is widely used represent many real-world systems. The collection and analysis of graph data using GNNs raise significant privacy concerns regarding disclosing sensitive information. Existing works in privacy-preserving ensure the nodes’ features labels. However, its structure also needs be privatized. To this problem, we provide a method LSPGNN that adds noise neighborhood node along with label. Here, perturb by sampling non-neighboring nodes randomizing them neighborhood. We use differentially private mechanisms graphs theoretical guarantees. This introduces challenge reducing impact on accuracy. In view, p -hop compensate for loss actual neighbors randomization. label as implemented previous methods GNNs. conduct extensive experiments datasets show perturbation structure. perform our proposed method.