面向表数据发布隐私保护的贪心聚类匿名方法

Greedy Clustering-Anonymity Method for Privacy Preservation of Table Data-Publishing

为了防范隐私泄露,表数据一般需要匿名处理后发布.现有匿名方案较少分类考察准标识属性概化,并缺少同时考虑信息损失量和时间效率的最优化.利用贪心法和聚类划分的思想,提出一种贪心聚类匿名方法:分类概化准标识属性,并分别度量其信息损失,有利于减小并合理评价信息损失.对元组间距离和元组与等价类距离,建立与最小合并概化信息损失值正相关的距离定义,聚类过程始终选取具有最小距离值的元组添加,从而保证信息损失总量趋于最小.按照k值控制逐一聚类,实现等价类均衡划分,减少了距离计算总量,节省了运行时间.实验结果表明,该方法在减少信息损失和运行时间方面是有效的.

To prevent privacy disclosure, table data generally needs to be anonymized before being published. Existing anonymity methods seldom distinguish different types of quasi-identifier in generalization, and also lack investigation into optimization of both information loss and time efficiency. In this paper, a greedy clustering-anonymity method is proposed using the ideas of greedy algorithm and clustering algorithm. The method makes distinct generalizations according to the type of quasi-identifier to conduct different calculations on information loss, and this providing reduction and reasonable estimate on information loss. Moreover, with regard to distance between tuples, or distance between a tuple and an equivalence class, two definitions are put forward in order to achieve minimum information loss in merging generalization. When establishing a new cluster, the tuple with the minimum distance in the ongoing cluster is always chosen to add. It ensures that the total information loss is close to minimum. Since the number of tuples in establishing each cluster is subject to k and the size of every cluster is equal to or just greater than k, the amount of calculation on distances and therefore the running time are reduced. Experimental results show that the proposed method is effective in reducing both information loss and running time.