Preserving privacy in data-publishing based on attribute weight and sensitivity rates

摘要

Publishing microdata while preserving individual privacy is a big concern for organizations. Generalization and Bucketization are two famous methods to handle this issue, but neither considers sensitivity levels for attributes and values. To describe sensitivity levels consider an example of a microdata with “Disease” and “Income” sensitive attributes where protecting “Disease” attribute is more important than “Income” attribute protection, also within Disease attribute values, protecting “cancer” value is more important than “cold” value protection. This paper presents a new technique based on attribute weighting for choosing the best method among generalization and bucketization when there are different significances for attributes protection. In addition, new diversity measure is introduced to handle different sensitivity rates in sensitive values; this is done by having special diversity for equivalence classes. Also “Sparse attack” for MSA microdata is introduced which is caused by unfavorable correlation of sensitive values, and a novel generalization technique for anonymize MSA microdata is presented.