基于聚类分析的个性化异构数据发布

由于异构数据的发布缺乏灵活性与实用性,提出了一种基于聚类分析的个性化异构数据发布方法.首先综合考虑数据的各种属性,通过聚类标签对数据的集群结构进行编码.另外,通过不断迭代更新原始数据能够始终保留数据的聚类结构,进一步在原始数据中加入噪声从而满足-差分隐私的要求.在满足差分隐私原则的前提下,提出了一种同时处理关系数据和集值数据的不确定性算法,不同类型的数据以类似的方式进行匿名化.通过实验验证了该方法能够有效提升异构数据发布的泛化能力,提升安全性与实用性.

Personalized Heterogeneous Data Publishing Based on Clustering Analysis

Aiming at the privacy protection of heterogeneous data publishing and the generalization of data mining, a differential privacy publishing scheme of heterogeneous data for clustering analysis was proposed. In order to solve the problem of lack of correct guidance after dealing with privacy information, the original data was grouped into clusters, and the cluster structure of data was coded by using the cluster label. In addition, a distance measurement cluster considering both relationship attribute and set-valued attribute was customized for heterogeneous data. Then, the original data was summarized iteratively while retaining the cluster structure. Furthermore, noise was added to the original data to meet the requirement of - differential privacy. On the premise of satisfying the principle of differential privacy, a uncertainty algorithm was proposed to process relational data and set-valued data simultaneously. Different types of data were anonymized in a similar way. Experiments show that this method can effectively solve the problem of heterogeneous data publishing.