Weak k-anonymity: a low-distortion model for protecting privacy

Sharing microdata tables is a primary concern in today information society. Privacy issues can be an obstacle to the free flow of such information. In recent years, disclosure control techniques have been developed to modify microdata tables in order to be anonymous. The k-anonymity framework has been widely adopted as a standard technique to remove links between public available identifiers (such as full names) and sensitive data contained in the shared tables. In this paper we give a weaker definition of k-anonymity, allowing lower distortion on the anonymized data. We show that, under the hypothesis in which the adversary is not sure a priori about the presence of a person in the table, the privacy properties of k-anonymity are respected also in the weak k-anonymity framework. Experiments on real-world data show that our approach outperforms k-anonymity in terms of distortion introduced in the released data by the algorithms to enforce anonymity.