基于差分隐私的异构数据聚类分析  

Heterogeneous data clustering analysis based on differential privacy

作　　者：雷美炼 曾路 张鑫鑫 沈博 LEI Meilian;ZENG Lu;ZHANG Xinxin;SHEN Bo(China Southern Power Grid Digital Grid Research Institute Co.LTD,Guangzhou 510663,China;State Key Laboratory of Information Security,Institute of Information Engineering,CAS,Beijing 100093,China)

机构地区：[1]南方电网数字电网研究院有限公司,广州510663 [2]中国科学院信息工程研究所信息安全国家重点实验室,北京100093

出　　处：《沈阳师范大学学报（自然科学版）》2022年第4期330-337,共8页Journal of Shenyang Normal University:Natural Science Edition

基　　金：国家自然科学基金资助项目(62172411,62172404)。

摘　　要：数据发布中的隐私保护问题是目前信息安全领域的研究热点之一。针对异构数据聚类分析场景下现有的差分隐私模型无法在隐私保护过程中有效保证数据完整性和可用性的问题,提出一种基于聚类分析的差分隐私异构数据发布方法。该方法将聚类分析问题转化为分类问题,利用类标签对原始数据的聚类结构同时进行泛化匿名机制和添加噪声处理后,发布满足ε-差分隐私保护的异构数据集。该方法实现了灵活的隐私保护聚类分析,同时保证了异构数据的隐私性和可用性。理论分析与实验结果表明,所提出的方法相比于同类算法具有更好的隐私性与可用性,在提高数据隐私保护处理效率的同时,保证了聚类分析结果的准确性。Privacy protection in data publishing is one of the research hotspots in the field of information security.Aiming at the problem that the existing differential privacy model can not effectively ensure the data integrity and availability in the process of privacy protection in the scenario of heterogeneous data cluster analysis,a differential privacy heterogeneous data publishing method based on cluster analysis is proposed.In this method,the clustering analysis problem is transformed into a classification problem,and the class label is used to generalize the clustering structure of the original data.After the anonymity mechanism is generalized and noise is added,the heterogeneous data sets satisfying-differential privacy protection are published.This method realizes flexible privacy protection cluster analysis,and ensures the privacy and availability of heterogeneous data.Theoretical analysis and experimental results show that compared with similar algorithms,the proposed method has better privacy and availability,improves the processing efficiency of data privacy protection and ensures the accuracy of clustering analysis results.

关 键 词：聚类分析 差分隐私 异构数据 匿名化 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]