基于极大关联属性集分解的隐私保护数据发布

传统的抽象化技术用于高维(属性)数据的匿名发布时会造成不可容忍的信息缺损,而分解技术虽然确保了数据真实性,但由于视图划分破坏了属性之间的内在关联,因此发布数据的可用性受到限制。该文提出了一种基于极大关联属性集的分解法MAAD(maximal associated attributes based decomposition),该方法利用频繁模式挖掘技术,寻找具有强关联性的属性组集合,并以此指导多视图的分解和生成。MAAD优先考虑了属性之间的关联性,所生成的多视图能够提供更好的数据挖掘性能。该文还定义了多视图发布的隐私保护模型-λmatching。实验结果表明:尤其在用于高维数据的匿名处理时,MAAD方法能够有效地提高数据可用性,具有很高的实用价值。

Maximum associate attribute based decomposition for privacy preserving publication

Privacy preserving data publication results in significant information loss caused by generalization when dealing with high-dimensional data.Decomposition improves the result by preserving exact values,but the released data still has low utility because the correlations between sensitive attributes and the other attributes are destroyed.A maximum associated attribute based decomposition model was developed to help determine the views for release.By using the frequent items mining technique,this model can find the attribute sets with strong associations.An anonymization principle λ-matching is used for multiple views publication.Tests demonstrate that the multiple views produced by this model are good for data mining since they preserve more correlations between the attributes than ordinary decomposition.