Randomization Methods to Ensure Data Privacy

Definition Many organizations, e.g., government statistical offices and search engine companies, collect potentially sensitive information regarding individuals either to publish this data for research, or in return for useful services. While some data collection organizations, like the census, are legally required not to breach the privacy of the individuals, other data collection organizations may not be trusted to uphold privacy. Hence, if U denotes the original data containing sensitive information about a set of individuals, then an untrusted data collector or researcher should only have access to an anonymized version of the data, U*, that does not disclose the sensitive information about the individuals. A randomized anonymization algorithm R is said to be a privacy preserving randomization method if for every table T, and for every output T * = R(T), the privacy of all the sensitive information of each individual in the original data is provably guaranteed.