关键词: 数据隐私保护, 属性聚类, 模式分解

Abstract: Most of the existing privacy preserving techniques often ignore special relation between sensitive attribute values and quasi-identifier attributes. At the same time, data privacy preserving need make anonymous publishing to meet composite privacy constraint for various field requirements. This paper proposes an efficient cluster algorithm based on model division for large data privacy preserving, by analyzing composite privacy constraint and similar sensitive attribute values. Firstly, it presents the clustering of sensitive attribute values to protect similar ones, and sets different weight to retain important quasi-identifier attributes. Secondly, the utility matrix of three-dimensional irregular matrix is used to obtain anonymous data with high accuracy and achieve the mode decomposition of anonymous data. Finally, experimental results on real data sets show that the data accurate rate and data error correction rate of the proposed algorithm obviously increase, and the approximate attack rate decreases.

Key words: data privacy preserving, attributes clustering, model division