Towards a privacy diagnosis centre: Measuring k-anonymity

Abstract

Most of the recent efforts addressing the issue of privacy have focused on devising algorithms for the anonymization and diversification of data. Our objective is upstream of these works: we are concerned with privacy diagnosis. In this paper, we start by investigating the issue of k-anonymity. We propose algorithms to explore various questions about k-anonymity of data. Such questions are, for instance, "is my data sufficiently anonymous?", "which information, if available from an outside source, threatens the anonymity of my data?" In this paper we focus on anonymity and, in particular, k-anonymity. The algorithms that we propose leverage two properties of k-anonymity that we express in the form of two lemmas. The first lemma is a monotonicity property that enables us to adapt the a-priori algorithm for k-anonymity. The second lemma is a determinism property that enables us to devise an efficient algorithm for δ-suppression. We illustrate and empirically analyze the performance of the proposed algorithms. © 2008 IEEE.