基于存在性隐私和关联性隐私的数据匿名技术

Data Anonymization Based on Presence and Association Privacy

在数据发布过程中有两种类型的隐私需要被保护:一是存在性隐私;一是关联性隐私.然而,现有的大部分面向数据发布中的隐私保护技术只关注二者之一.因此,提出一种新的技术:MBOA,它的思想是:将原始数据中的准标识符属性和敏感属性分成两个不同的表,设定一个概率α,在存在性概率不超过阈值α的前提下,遵循尽量减少信息损失的原则,在准标识符表中插入额外的元组.进一步,本文将MBOA技术与l-多样性模型结合,提出了(α,l)-多样性模型,并设计了一个用来构造满足(α,l)-多样性的MBOA模式的算法,理论分析和实验表明该方法既能保护存在性隐私和关联性隐私,又具有较小的信息损失.

There are two kinds of privacy that must be preserved： presence privacy and association privacy. However, most of the ex- isting privacy-preserving data publishing techniques only focus on either of them. Therefore, in this paper, we propose a novel tech- nique ： MBOA. It divides all the quasi-identifier and sensitive values of the raw data into two separated tables, then setting α probabil- ity value α, inserting some additional tuples into quasi-identifier table according to a mechanism while the presence probability is not more than α. Furthermore, we combine MBOA with /-diversity and propose （ α,l） -diversity. We develop an algorithm of MBOA scheme that provides sufficient protection of both presence privacy and association privacy with low information loss and satisfies （α,l） -diversity.