基于置信度分析的差分隐私保护参数配置方法研究

Research on differential privacy protection parameter configuration method based on confidence level

为了解决数据发布和分析过程中用户真实数据信息被披露的问题,降低攻击者通过差分攻击和概率推理攻击获取真实结果的概率,提出了一种基于置信度分析的差分隐私保护参数配置方法。在攻击者概率推理攻击模型下对攻击者置信度进行分析,使之不高于根据数据隐私属性所设置的隐私概率阈值。所提出的方法能够针对不同查询用户查询权限的差异配置更加合理的隐私保护参数,避免了隐私披露的风险。实验分析表明,所提出的方法根据查询权限、噪声分布特性以及数据隐私属性分析攻击者置信度与隐私保护参数的对应关系,并据此推导出隐私保护参数的配置公式,从而在不违背隐私保护概率阈值的情况下配置合适的ε参数。

In order to solve the problem that the user's real data information is disclosed during the data release and analysis process, and reduce the probability of an attacker gaining real results through differential attacks and probabilistic inference attacks, a differential privacy protection parameter configuration method based on confidence level is proposed. Analysis of attacker confidence under attacker probabilistic inference attack model and make it no higher than the privacy probability threshold set according to the data privacy attribute. The proposed method can configure more reasonable privacy protection parameters for different query privilege of query users, and avoids the risk of privacy disclosure. The experimental analysis shows that the proposed method analyzes the correspondence between attacker confidence level and privacy protection parameters based on query privilege, noise distribution characteristics and data privacy attributes, and derives the configuration formula of privacy protection parameters, which configure the appropriate parameters without violating the privacy protection probability threshold.