【摘要】 针对面向聚类应用的隐私保护数据发布问题,引入邻域相关性概念,提出了一种基于邻域相关性的数据扰动算法NCDP,分析每个数据点邻域中与其邻域亲密的所有点以及邻域的平衡性,在不平衡情况下除去亲密集中可能的局部噪声数据点,向每个邻域亲密点进行一定长度的平移,得到扰动后的数据点.理论分析表明,扰动后的数据点不仅实现了对原始数值的保护,而且扰动前后数据点的邻域亲密点仍然维持亲密关系,从而保持了邻域的稳定性.实验采用k-means和DBSCAN聚类算法对扰动前后的数据进行聚类,并且与其他扰动算法进行了分析对比.实验结果表明,算法NCDP扰动前后的数据聚类结果有较高的相似度,可以较好地兼顾保护数据隐私与维持聚类可用性.更多还原

【Abstract】 Concerning the problem of privacy preserving data publishing for clustering and introducing the concept of neighborhood correlation,a novel privacy preserving data perturbation NCDP based on neighborhood correlation is proposed in the paper.NCDP analyzes a point’s intimate data points and the balance of the point’s neighborhood.Furthermore,NCDP perturbs the data point by removing all the local noise points and shifting point towards to each intimate point in a certain distance when the neighborhood is unbalanced.Theoretical analysis testifies that this perturbation strategy not only achieves the protection of the original data but also maintains the intimate relationship between the perturbed data point and intimate points.Experimental analysis is designed by adopting clustering algorithm k-means and DBSCAN on primitive datasets and perturbed ones and comparing experimental results with other perturbation algorithms.Experimental results prove that the algorithm NCDP has a higher similar clustering result with primitive datasets and can preserve the data privacy meanwhile maintaining better clustering usability.更多还原