Privacy Risk Against Composition Attack

Abstract

Privacy in multiple independent data publishing has attracted considerable research interest in recent years. Although each published data set poses a small privacy risk to individuals, recent studies show that this risk increases when different organizations have some common records and they publish their data sets independently without any coordination with each other. If an individual can be detected from disparate providers, the individual's privacy is compromised. This type of privacy breach is called composition attack. A few studies have done to mitigate this attack. However, none of them studies the risk against this attack from a single data set. Motivated by this gap, this paper uses a probabilistic model to estimate the risk against composition attack from a single data. Therefore, a publisher can predict the risk against composition attack of a data set prior to publication. To evaluate the effectiveness of our model we also perform empirical analysis to show that the estimated risk can give us the pattern of the real risk.