Privacy-preserving model and generalization correlation attacks for 1:M data with multiple sensitive attributes

Abstract

Preserved privacy and enhanced utility are two competing requirements in data publishing. For maintaining a trade-off between the two; a plethora of research work exist in 1:1 scenario (each individual has a single record) with a single sensitive attribute (SA). However, some practical scenarios i.e., data having 1:M records (an individual can have multiple records) with multiple sensitive attributes (MSAs), have been relatively understudied. In our current interconnected and digitalized society, the capability to deal with such scenarios is increasingly important due to the ever-increasing sources of data that could be drawn together and infer one's information (e.g. profile and lifestyle) and consequently, compromising one's privacy. In this paper, we present a new type of attack on 1:M records with MSAs, coined as MSAs generalization correlation attacks and perform formal modeling and analysis of these attacks. Then, we propose a privacy-preserving technique “(p, l)-Angelization” for 1:M–MSAs data publication. Extensive experiments over real-world datasets advocate the outperformance of our technique over its counterparts.