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.