Stipulation-Based Anonymization with Sensitivity Flags for Privacy Preserving Data Publishing

摘要

Privacy is a major concern for organizations that release Microdata for informal analysis. Most of the Privacy Preserving Data Publishing (PPDP) techniques anonymize data based on personalized privacy requirements or based on some general utility specification. The consequence is that, either the record owner's privacy requirements or the data miner's (analyst's) data efficacy requirements are considered for data anonymization, which leads to tainted accuracy in several data mining tasks. Motivated by this we propose a novel approach which considers privacy requirements in the form of Sensitivity Flags from the record owners end, as well as Application Specific Requirements from the data miners (analysts) end. Our proposed method is theoretically analyzed and the mathematical analysis outperforms the earlier works with sufficient experiments.