Local perturbation of critical subgraph based social network structural privacy protection

Abstract

In social networks, the direct exposure of structural data containing numerous authentic relationships significantly risks user privacy leakage. Thus, safeguarding the privacy of such data has emerged as a pressing concern. Existing methodologies commonly employ global perturbation, randomly altering the network structure and resulting in diminished data availability upon publication. Notably, these methods often overlook the privacy implications associated with critical nodes and diverse structural features within subgraphs, thereby exacerbating the risk of privacy breaches. In response to these challenges, we present a novel privacy protection approach, termed CGLP, tailored for social network structures and centered on the concept of local perturbation of critical subgraphs. Specifically, CGLP meticulously identifies critical nodes, factoring in local structural complexity and positional features, and subsequently identifies subcritical nodes exhibiting structural similarity through graph embedding and clustering. The method further constructs critical node subgraph and core critical node subgraphs. Crucially, diverse local perturbation methods are strategically designed for subgraphs with varying structural features, mitigating the potential issues of under-protection and over-protection. Evaluation results on real datasets demonstrate that CGLP effectively balances the availability and privacy of the published network structure, concurrently addressing the privacy concerns associated with critical nodes.