DCKPDP：改进k-prototype聚类的差分隐私混合属性数据发布方法

当前混合属性数据发布中隐私保护方法大多存在隐私保护效果不佳或数据效用较差的问题,采用差分隐私与优化的k-prototype聚类方法相结合,提出改进k-prototype聚类的差分隐私混合属性数据发布方法(DCKPDP)。为解决传统k-prototype聚类算法没有考虑不同数值型属性对聚类结果有较大影响的问题,利用信息熵为每个数值型属性添加属性权重;为解决聚类初始中心点人为规定或者由随机算法随机确定,导致聚类结果精确度不高的问题,结合数据对象的局部密度和高密度对聚类过程中初始中心点进行自适应选择;为解决数据信息泄露风险较高的问题,对聚类中心值进行差分隐私保护。实验结果表明,DCKPDP算法满足差分隐私保护所需的噪声量更小,数据的可用性更好。

Differential privacy mixed attribute data publishing method for improved k-prototype clustering

Most of the current privacy protection methods in mixed attribute data publishing have problems of poor privacy protection effect or poor data utility. This paper proposed a differential privacy mixed attribute data publishing method(DCKPDP) based on improved k-prototype clustering. In order to solve the problem that the traditional k-prototype clustering algorithm did not consider the great influence of different numerical attributes on clustering results, using the information entropy added weight for each numerical attribute. In order to solve the problem of low accuracy of clustering results caused by artificial initial center points or randomly determined by random algorithm, combining the local density and high density of data objects carried out an adaptive selection of initial center points in the process of clustering. In order to solve the problem of high risk of data information leakage, using differential privacy protected the clustering center value. The experimental results show that the DCKPDP algorithm satisfies the requirement of differential privacy protection with less noise and better data availability.