Personalized anonymity privacy protection algorithm based on clustering

• 1、Computer and Information Technology School, Beijing Jiaotong University, Beijing 100044

Abstract：Anonymous privacy protection technology is the most widely used data privacy protection technology. Its principle is to publish semantically consistent data by generalizing or hiding the quasi-identifier attributePersonalized in the original data table. Most anonymity models for privacy preserving neither consider the semantic similarity of sensitive attribute values, which are vulnerable to similarity attack, nor achieve a balance between data privacy and availability. This paper proposes a (a,k,d)-anonymity privacy preserving algorithm based on clustering. The algorithm defines semantic similarity groups for sensitive attributes, and requires that the number of semantic similarity groups in each equivalence class is not less than d to prevent similarity attack. In addition, it satisfies the personalized needs of the anonymity model by setting different frequency constraints for sensitive attributes that differ in the equivalence class to limit their frequency. The anonymity algorithm is implemented by combining the maximum dissimilarity clustering, which improves the availability of anonymous data on the basis of privacy. The experimental results show that this algorithm can reduce the amount of information loss by more than 50%, resist the similarity attack and provide personalized privacy protection with less time cost than other k-anonymity models based on clustering.

Keywords： privacy protection similarity attack clustering personalization