Improving the Utility of Differentially Private Data Releases via k-Anonymity

摘要

A common view in some data anonymization literature is to oppose the "old'' k-anonymity model to the "new'' differential privacy model, which offers more robust privacy guarantees. However, the utility of the masked results provided by differential privacy is usually limited, due to the amount of noise that needs to be added to the output, or because utility can only be guaranteed for a restricted type of queries. This is in contrast with the general-purpose anonymized data resulting from k-anonymity mechanisms, which also focus on preserving data utility. In this paper, we show that a synergy between differential privacy and k-anonymity can be found when the objective is to release anonymized data: k-anonymity can help improving the utility of the differentially private release. Specifically, we show that the amount of noise required to fulfill differential privacy can be reduced if noise is added to a k-anonymous version of the data set, where k-anonymity is reached through a specially designed microaggregation of all attributes. As a result of noise reduction, the analytical utility of the anonymized output data set is increased. The theoretical benefits of our proposal are illustrated in a practical setting with an empirical evaluation on a reference data set.