核子空间投影和广义特征值分解的云数据隐私保护

CLOUD DATA PRIVACY PROTECTION BASED ON KERNEL SUBSPACE PROJECTION AND GENERALIZED EIGENVALUE DECOMPOSITION

云数据及大量网络数据需要在保护用户隐私的前提下进行统计和分析,对此提出核子空间投影和广义特征值分解的云数据隐私保护算法。将云数据进行数学化建模,该算法将隐私保护作为数据转换问题进行处理,转换问题分为隐私不敏感任务和隐私敏感任务两类。该算法形成类间散布矩阵,寻找子空间维度来解决特征值分解问题,并对广义特征值排序,得到广义特征向量对应的最大广义特征值;对云数据进行转换,实现数据隐私保护。实验结果表明,该方法能够实现用户隐私保护,并且核广义特征值分解算法优于子空间隐私保护算法,且两种隐私保护算法都优于其他隐私保护方法。

Cloud data and a large number of network data need to be counted and analyzed under the premise of protecting user privacy. We proposed cloud data privacy protection algorithm based on kernel subspace projection and generalized eigenvalue decomposition. The cloud data was modeled mathematically. The privacy protection was treated as data conversion problem, which could be divided into two categories: privacy insensitive task and privacy sensitive task. The algorithm formed a scatter matrix between classes, found subspace dimensions to solve eigenvalue decomposition, and ranked the generalized eigenvalues to get the maximum generalized eigenvalues corresponding to the generalized eigenvectors. The cloud data was transformed to realize data privacy protection. The experimental results show that the proposed method can achieve user privacy protection, and the kernel generalized eigenvalue decomposition algorithm is superior to the subspace privacy protection algorithm. And the two privacy protection algorithms are superior to other privacy protection methods.