Anonymity: A Formalization of Privacy - `-Diversity

Anonymization of published microdata has become a very important topic nowadays. The major difficulty is to publish data of individuals in a manner that the released table both provides enough information to the public and prevents disclosure of sensitive information. Therefore, several authors proposed definitions of privacy to get anonymous microdata. One definition is called k-Anonymity and states that every individual in one generalized block is indistinguishable from at least k 1 other individuals. `-Diversity uses a stronger privacy definition and claims that every generalized block has to contain at least ` different sensitive values. Another definition is called t-Closeness. It demands that the distribution of one sensitive value of a generalized block is close to its distribution in the entire table. This paper mainly deals with the principle and notion of `Diversity. Therefore, two methods called Homogeneity and Background-Knowledge Attack are discussed to break the privacy constraints of k-Anonymity. Then a model to reason about privacy in microdata, namely Bayes-Optimal Privacy, is introduced. Based on k-Anonymity and Bayes-Optimal Privacy the principle and several instantiations of `-Diversity are discussed. At the end `-Diversity is applied to a real database gathered from several Android devices.