作者:韩启龙，吴晓明
Authors:HANQilong,WUXiaoming摘要:针对工业企业图数据链接预测过程中 ,节点间的敏感数据面临隐私泄露的问题 , 以本地差分隐私理论 为基础 ,从链接预测任务表现的角度分析了现有的图数据建模方法在隐私保护上的缺点和不足 。提出在个性化采 样技术的随机响应机制 ,减少用户端噪声添加 , 同时结合两轮数据收集的子图划分策略 ,保留原始图数据中子图聚 集特征 ,最终实现了 一种个性化采样随机响应本地差分隐私( Personalized Sampling Randomized Response Local Dif- ferential Privacy ,PSRR-LDP) 图数据建模算法 ,理论证明 PSRR-LDP 算法满足 ε -边本地差分隐私 。仿真实验结果表 明 ,PSRR-LDP 算法在保证隐私的同时具有更优的链接预测效果。
Abstract:To solve the problem of node sensitive link privacy being exposed in the process of link prediction on industrial business graph data , according to the theory of local differential privacy , the shortcomings of the existing graph privacy protection technology are analyzed from the perspective of link prediction task performance. Based on the existing randomized response mechanism , it introduces the personalized sampling technology to reduce the noise addition on the user side. At the same time , combined with the subgraph partitioning strategy of two rounds of data collection , the subgraph cluster feature of the original graph is retained. Finally , a personalized sampling randomized response local differential privacy ( PSRR-LDP) graph data perturbing algorithm was implemented , and the PSRR-LDP algorithm is theoretically proved to satisfy the ε -edge Local differential privacy. The simulation experiments show that the PSRR-LDP algorithm has better link prediction performance while ensuring privacy.

PDF全文下载地址:

可免费Download/下载PDF全文