基于联合树的高维数据本地化差分隐私保护算法  

Local differential privacy protection algorithm for high dimensional data based on junction tree

作　　者：程思源 龙士工[1,2] CHENG Si-yuan;LONG Shi-gong(State Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025,China;College of Computer Science and Technology,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学公共大数据国家重点实验室,贵州贵阳550025 [2]贵州大学计算机科学与技术学院,贵州贵阳550025

出　　处：《计算机工程与设计》2024年第6期1601-1606,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(62062020)。

摘　　要：为解决发布高维数据过程中复杂的属性关联问题并避免中心服务器不可信任的问题,提出一种基于联合树的高维数据本地化差分隐私保护算法(JT-LDP算法)。基于不可信的中心服务器实现对用户数据的本地化差分隐私保护,中心服务器接收到用户本地化差分隐私保护的数据后,基于联合树算法识别高维数据的属性相关性,将高维数据属性集分割成多个独立的低维属性集。通过采样合成新的数据集进行发布。实验结果表明,JT-LDP算法在高维数据情况下具有更高的精度。To solve the complex attribute association problem in the process of publishing high-dimensional data and avoid the untrustworthy problem of the central server,a local differential privacy protection algorithm(JT-LDP algorithm)of high-dimensional data based on junction tree was proposed.The local differential privacy protection of users’data was implemented based on the untrusted central server.After receiving the data protected by local differential privacy,the central server recognized the attribute correlation of the high dimensional data based on the junction tree algorithm,and the high dimensional data attribute set was divided into multiple independent low dimensional attribute sets.New data sets were synthesized by sampling and published.Experimental results show that JT-LDP algorithm has higher accuracy in the case of high dimensional data.

关 键 词：高维数据 本地化差分隐私 联合树 数据发布 联合分布估计 马尔可夫网 随机响应 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]