面向聚类挖掘的局部旋转扰动隐私保护算法

Partial Rotation Perturbation for Privacy-Preserving Clustering Mining

聚类挖掘可以高效准确地从数据中找出很多潜在的、有价值的规律,但也同时存在着泄露用户隐私数据的安全威胁.已经有一些专门针对聚类挖掘的隐私保护研究,其中乘法扰动方法是一种准确性和安全性都较高的隐私保护算法.研究发现已知信息独立分量分析极大地降低了已有乘法扰动方法的安全性,它能够从乘法扰动数据中近似估计隐私数据.为了解决以上问题,提出了局部旋转扰动隐私保护算法,通过准确性分析得出新算法具有零损失准确性.利用安全性分析证明新算法能够有效抵御独立分量分析的攻击,具有更高的安全性.将新算法应用到聚类挖掘中,得到了与未加隐私保护的聚类挖掘非常接近的结果,说明了它的可行性.局部旋转扰动方法的出现,有效地解决了已有乘法扰动方法的安全漏洞,使得聚类挖掘能够更加安全地得到应用.

Many potential and valuable rules can be derived from data via clustering mining in an effective and accurate way, which can lead to security threats such as the disclosure of user privacy. Many privacy-preserving researches on clustering mining have been conducted, especially on muhiplicative perturbation （MP） that is a highly secure and accurate method. Research finds known knowledge independ- ent component analysis （KK-ICA） can greatly reduce the privacy security of existing MP. It can approximately estimate private data from MP data. To solve the problem, partial rotation perturbation （PRP） is proposed. The analysis of accuracy shows that PRP has zero-loss accuracy. The analysis of security proves that PRP can defend attack from the KK-ICA availably and is more secure. PRP is applied in clustering mining. The results are very similar to unpreserved clustering mining results, which shows that MP is practicable. The existence of PRP solves the problem with the security vulnerability of existing MP effectively, making the application of clustering more secure.