Disclosure Detection in Multivariate Categorical Databases: Auditing Confidentiality Protection Through Two New Matrix Operators
针对多维分类数据库中推断性披露问题，提出一种数据值依赖的方法，通过两种新矩阵算子实现高效计算，为数据库管理员提供审计保密性保护的工具。
As databases grow more prevalent and comprehensive, database administrators seek to limit disclosure of confidential information while still providing access to data. Practical databases accommodate users with heterogeneous needs for access. Each class of data user is accorded access to only certain views. Other views are considered confidential, and hence to be protected. Using illustrations from health care and education, this article addresses inferential disclosure of confidential views in multidimensional categorical databases. It demonstrates that any structural, so data-value-independent method for detecting disclosure can fail. Consistent with previous work for two-way tables, it presents a data-value-dependent method to obtain tight lower and upper bounds for confidential data values. For two-dimensional projections of categorical databases, it exploits the network structure of a linear programming (LP) formulation to develop two transportation flow algorithms that are both computationally efficient and insightful. These algorithms can be easily implemented through two new matrix operators, cell-maxima and cell-minima. Collectively, this method is called matrix comparative assignment (MCA). Finally, it extends both the LP and MCA approaches to inferential disclosure when accessible views have been masked.