SaDes: An Interactive System for Sensitivity-aware Desensitization towards Tabular Data

Kechun Zhao, Hui Li, Zheng Gong, Jiangtao Cui

Abstract: Before the publication of particular datasets, in order to protect the private information while preserving the usability as much as possible, desensitization is required. Automatic identification and evaluation of sensitive attributes are prerequisites for targeted desensitization of datasets, sensitivity can also reflect the effect of desensitization in turn. However, existing desensitization systems all rely on predefined desensitization model with respect to manually given sensitivity levels, which is subjective and unable to be applied end-to-end. Besides, there is no way for the user to tell whether the desensitization is performed enough or superfluous. In this demonstration, we present an interactive system for sensitivity-aware desensitization towards tabular data (SaDes). It automatically evaluates the risks of re-identification for arbitrary columns according to record-linkage attack, and performs desensitization accordingly. The risks of re-identification for the desensitized data can be immediately evaluated such that the user can iteratively execute desensitization in order to achieve a better balance between the usability and privacy. To the best of our knowledge, SaDes is the first system that provides automatic sensitivity evaluation and interactive desensitization in a back-to-back manner.