A Game Theoretic Approach Applied in k- Anonymization for Preserving Privacy in Shared Data

Abstract

Privacy preservation is one of the greatest concerns when data is shared between different organizations. On the one hand, releasing data for research purposes is inevitable. On the other hand, sharing this data can jeopardize users' privacy. An effective solution, for the sharing organizations, is to use anonymization techniques to hide the users' sensitive information. One of the most popular anonymization techniques is k-Anonymization in which any data record is indistinguishable from at least k-1 other records. However, one of the fundamental challenges in choosing the value of k is the trade-off between achieving a higher privacy and the information loss associated with the anonymization. In this work, the problem of choosing the optimal anonymization level for k-anonymization, under possible attacks, is studied when multiple organizations share their data to a common platform which is data collector (Cybex) in this case. In particular, we have considered two common types of attacks, namely, Homogeneity attack and Background knowledge attack, which have the capability of compromising k-anonymization technique. To this end, a novel game-theoretic framework is proposed to model the interactions between the sharing organizations and the attacker along with contract theoretic framework to model interactions between organizations and data collector (Cybex). The problem is first formulated as a static game and its different Nash equilibria solutions are analytically derived. Later, we have used a contract theoretic model on interactions between data collector (Cybex) and the organizations. We also show how data collector varies the rewards of the organizations to increase it's utility over the stages.