基于个体相关性的隐私保护方法研究

发布时间：2018-03-27 23:29

本文选题：隐私保护　切入点：背景知识攻击　出处：《华中科技大学》2016年硕士论文

【摘要】：随着各种社交网络、个性化推荐等服务的发展,个人信息往往被服务提供者收集、管理并加以利用,由此也产生了个人信息被泄漏的风险。现有的个人信息隐私保护方法,在具有各种背景知识的攻击者面前面临着更加严峻的挑战。因此,研究和改进隐私保护方法以适应新的攻击场景具有重要意义。针对目前的隐私保护方法对已知个体相关性进行攻击的场景研究不足的问题,以一种比较常见的已知个体相关性的攻击场景为背景,设计了一种能够抵抗该攻击的隐私保护模型,即r-抗元组关系攻击隐私保护模型。为了得到该隐私保护模型,首先提取出该攻击场景的本质并抽象成已知元组关系的攻击模型,并对攻击模型中的个体间关系进行范围的界定和建模;然后给出能够抵抗该攻击模型的元组应该满足的条件即r-抗元组关系攻击性,约束匿名后的关联元组的候选敏感属性集合之间的交集大小至少为阈值r;最后根据是否包含关联元组为不同类型的分组分别施加不同程度的隐私约束,给出抗元组关系攻击隐私保护模型的定义,并从理论上证明模型的安全性。以r-抗元组关系攻击隐私保护模型为基础,设计出用于生成匿名数据集的算法,包括用于提取数据集中背景知识的敏感属性等值关系提取算法以及用于生成满足安全约束的匿名发布表的抗元组关系攻击隐私保护算法(包括分组创建算法、分组补充算法和表分割三个部分),并给出算法正确性、安全性、可用性以及代价的理论分析。实验表明,满足r-抗元组关系攻击隐私保护模型的隐私保护算法生成的匿名数据与满足?-多样性的Anatomy算法生成的匿名数据相比,两者不仅具有相近的可用性,而且前者具有更好的安全性。
[Abstract]:With the development of various social networks, personalized recommendation and other services, personal information is often collected, managed and utilized by service providers. There are even more serious challenges facing attackers with a variety of backgrounds. It is of great significance to study and improve privacy protection methods to adapt to new attack scenarios. Based on a common attack scenario with known individual correlation, this paper designs a privacy protection model that can resist this attack, namely r-tuple relation attack privacy protection model, in order to obtain the privacy protection model. Firstly, the essence of the attack scene is extracted and abstracted into an attack model with known tuple relationships, and the scope of the relationship between individuals in the attack model is defined and modeled. Then, the condition that the tuple can resist the attack model is given, that is, r-anti-tuple relation aggression. The size of the intersection between candidate sensitive attribute sets after constrained anonymous tuples is at least a threshold r; finally, privacy constraints are imposed to varying degrees depending on whether groups containing association tuples are of different types. This paper gives the definition of privacy protection model against tuple relation attack, and proves the security of the model theoretically. Based on the privacy protection model of r-tuple relation attack, an algorithm is designed to generate anonymous dataset. It includes a sensitive attribute equivalence extraction algorithm for extracting background knowledge in a dataset and an anti-tuple relational attack privacy protection algorithm for generating anonymous publishing tables that meet security constraints (including grouping creation algorithm). The theoretical analysis of the correctness, security, availability and cost of the algorithm is given. The experimental results show that, The anonymous data and satisfaction generated by privacy protection algorithm satisfying r-tuple relation attack privacy protection model? Compared with the anonymous data generated by the diversity Anatomy algorithm, the two methods not only have similar availability, but also have better security.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP309

【参考文献】