基于本地差分隐私的K-modes聚类数据隐私保护方法  

Privacy Protection Method for K-modes Clustering Data with Local Dif⁃ferential Privacy

作　　者：张少波[1,2] 原刘杰 毛新军 朱更明[1] ZHANG Shao-bo;YUAN Liu-jie;MAO Xin-jun;ZHU Geng-ming(School of Computer Science and Engineering,Hunan University of Science and Technology,Xiangtan,Hunan 411201,China;Key Laboratory of Software Engineering for Complex Systems,National University of Defense Technology,Changsha,Hunan 410073,China)

机构地区：[1]湖南科技大学计算机科学与工程学院,湖南湘潭411201 [2]国防科技大学复杂系统软件工程重点实验室,湖南长沙410073

出　　处：《电子学报》2022年第9期2181-2188,共8页Acta Electronica Sinica

基　　金：湖南省自然科学基金面上项目(No.2020JJ4317,No.2020JJ4250);湖南省教育厅科学研究重点项目(No.21A0318,No.19A275);湖南省研究生科研创新项目(No.CX20200999)。

摘　　要：分类型数据聚类是数据挖掘的重要研究内容,聚类数据中通常包含用户一些敏感信息.为保护聚类数据中的用户隐私,当前主要采用基于可信第三方隐私保护模型,但现实中第三方也存在隐私泄露风险.针对此问题,该文引入本地差分隐私技术,提出一种去可信第三方的K-modes聚类数据隐私保护方法.该方法首先利用随机采样技术对数据进行采样,然后使用本地差分隐私技术对采样数据进行扰动,最后通过聚类服务端与用户的交互迭代完成聚类.在聚类过程中,无需可信第三方对数据进行隐私预处理,避免了第三方泄露用户隐私的风险.理论分析证明了该方法的隐私性和可行性,实验结果表明该方法在满足本地差分隐私机制的前提下保证了聚类结果的质量.Categorical data clustering is an important research content for data mining,and clustering data usually contains some sensitive information of user.In order to protect user privacy in clustering data,the privacy protection model based on trusted third-party is currently mainly adopted.However,in reality,the third-party also has the risk of privacy leak⁃age.In this paper,we propose a privacy protection method for K-modes clustering data without trusted third-party by intro⁃ducing local differential privacy technology.Our method first uses random sampling technology to sample the data,then perturbs the sampled data by using local differential privacy technology,and finally complete the clustering through the in⁃teraction between the server and the user.In the clustering process,our method does not require a trusted third-party to per⁃form privacy preprocessing on the data,which avoids the risk of the third-party disclosing the user's privacy.Theoretical analysis proves the privacy and feasibility of our method.Experimental results show that our method guarantees the quality of the clustering results under the premise of satisfying the local differential privacy mechanism.

关 键 词：隐私保护 本地差分隐私 数据挖掘 K-modes聚类 去可信第三方 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]