A Graph Generation Network with Privacy Preserving Capabilities

The use of large datasets for data mining and analysis can stimulate progress in science and technology while also propelling economic growth. Graph-structured data is a crucial component of both data mining and analysis. However, this type of data often contains sensitive personal information, making it vulnerable to potential attacks and widespread privacy breaches. Graph data encodes sensitive information, including personal attributes (nodes) and complex interaction relationships (edges). Rényi differential privacy provides a stricter definition of privacy protection. This paper introduces the RDP-GGAN framework, which integrates Rényi differential privacy technology with generative adversarial networks to offer improved privacy protection capabilities. The framework utilizes Rényi differential privacy to establish and enforce strict privacy constraints for deep graph generative models, with a particular emphasis on preserving edge privacy in graph data to ensure connection privacy in relational data. To enhance edge differential privacy, appropriate noise is injected into the gradient of link-reconstruction-based graph generative models.