一种基于特征值分解的数据挖掘隐私保护扰乱增强方法

An SVD-Based Advanced Data Perturbation Method for Privacy-Preserving Data Mining

针对数据扰乱技术中的特征值分解攻击方法,分析和评估了该攻击下数据扰乱模型的安全性,发现现有模型存在一定的脆弱性.设计了基于特征空间的扰乱强度量化方法,对隐私保护强度进行量化评估.在此基础上,通过阀值曲线的上限投影,提出了针对数据分离攻击的隐私保护的增强方法.结果表明,在盲数据源下,该增强方法对于抵抗特征值分解攻击具有有效性和鲁棒性.

Analysis and evaluations on the security of general data perturbation method through SVD （Singular Vector Decomposition） based attacks shows its vulnerability. An evaluation method was raised for quantifying the strength of data perturbation, and an enhanced method was presented based on eigen space to prevent against SVD-based attacks. The experiments show the availability and robustness of the model with unknown data source under SVD attacks.