Towards the Diversity of Sensitive Attributes in k-Anonymity

Abstract

Privacy preservation is an important and challenging problem in microdata release. As a de-identification model, k-anonymity has gained much attention recently. While focusing on identity disclosures, k-anonymity does not well resolve attribute disclosures. In this paper we focus on the sensitive attribute disclosures in k-anonymity and propose an ordinal distance based sensitivity aware diversity metric. We assume the more diversity the sensitive attribute assumes in an equivalence class in a k-anonymized table, the less inference channel there is in the equivalence class