A Decision Tree Approach for Assessing and Mitigating Background and Identity Disclosure Risks

The Facebook/Cambridge Analytica data scandal shows a type of privacy threat where an adversary attacks on a massive number of people without prior knowledge about their background information. Existing studies typically assume that the adversary knew the background information of the target individuals. This study examines the disclosure risk issue in privacy breaches without such an assumption. We define the background disclosure risk and re-identification risk based on the notion of prior and conditional probabilities respectively, and integrate the two risk measures into a composite measure using the Minimum Description Length principle. We then develop a decision-tree pruning algorithm to find an appropriate group size considering the tradeoff between disclosure risk and data utility. Furthermore, we propose a novel tiered generalization method for anonymizing data at the group level. An experimental study has been conducted to demonstrate the effectiveness of our approach