(t, λ)-Uniqueness: Anonymity Management for Data Publication

Recent work has shown that the adversary\´s background knowledge is a very important factor in privacy-preserving data publishing. In this paper, we formalize background knowledge h of form "an individual X\´s sensitive value belongs to class C or range W. Through analyzing the drawbacks of previous approaches in dealing with this form of background knowledge, we develop a novel privacy criterion (tau, lambda)-uniqueness that sufficiently defends against attacks leveraging the background knowledge h. We accompany the criterion with an effective algorithm, which computes a privacy-guarded published table that permits retrieval of accurate aggregate information about the micro-data. We illustrate its advantages through theoretical analysis and experimental validation.