Social influence prediction method and system based on heterogeneous graph neural network privacy protection

• Main Authors: YAN DONGQI, WANG JINYAN, WEI YUECEN, LI XIANXIAN, FU XINGCHENG
• Format: Patent
• Language: Chinese; English
• Published: 04.10.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a social influence prediction method based on heterogeneous graph neural network privacy protection, and the method comprises the following steps: 1, defining a meta-path representing semantic information in a heterogeneous graph, and generating a multi-association information protocol on the heterogeneous graph into a same composition according to meta-path migration; 2, attention sensitivity is introduced based on differential privacy to generate Gaussian noise, and the noise and a node learning result are aggregated into node embedding; 3, using the protected node features and the original graph structure for structural privacy protection learning of the heterogeneous graph, and generating a relation sub-graph for each type of edge; 4, setting a gradient and a cutting boundary of topological learning of the perturbation graph as a gradient mean value; 5, distributing privacy protection strength by using a double-layer optimization mechanism to guarantee the social influence predicti