XColor: Protecting General Proximity Privacy

As a severe threat in anonymized data publication, proximity breach is gaining increasing attention. Such breach occurs when an attacker learns with high confidence that the sensitive value of a targeted victim belongs to a set of semantically proximate values, even though not sure about the exact value. Focusing on tackling proximity breach under specific data models (e.g., categorical or numeric sensitive data), existing research efforts fail to address the threat for a much richer set of models wherein the semantic proximity might be defined by arbitrary functions. This work advocates studying proximity privacy in a data-model-neutral manner, thereby providing solutions of general applicability. Specifically, a) we define the association between quasi-identifier and sensitive attributes (QI-SA association) under a highly abstract data model; b) we formalize proximity breach in the general framework of proximate QISA association, and remedy such breach with a unified privacy definition; c) for the proposed definition, we conduct an analytical study on its characteristics, and derive criteria to efficiently test its satisfiability; d) further, we devise a novel anonymization model, XCOLOR, to fulfill this definition, with guarantees on both operation efficiency and utility preservation. A comprehensive empirical evaluation is performed to validate the analytical models and the efficacy of the XCOLOR method.