Towards a More Realistic Disclosure Risk Assessment

摘要

The score was introduced in 2001 in order to compare different perturbative methods for statistical database protection. It measures the trade-off between utility (information loss) and privacy (disclosure risk of the released data). Since its introduction, the score has been widely accepted and used in the statistical database community. In particular, some methods are sometimes prefered to others depending on the obtained results in the original computation of the score. In this paper we argue that some original aspects of the score computation, specially those related to the disclosure risk, should be revisited. Informally, the reason is that they do not consider the best possible situation for the intruder, and so they do not measure the real level of privacy. We add some experimental results which support our claims. More importantly, we propose some modifications which can/should lead in the future to a more fair, realistic and useful computation of the score.