Checking And Preventing Privacy Inference Attacks Based On K-Anonymized Microdata

K-Anonymity is proposed as the prospective microdata publication model, which also faces the challenges that how to grant the anonymized microdata the inference-proof ability in diverse environment. Study on strong anti-inference k-anonymmity model is attracting more and more attentions from theoretical and industrial areas.The purpose of the paper is to detect and prevent privacy inference attacks on k-anonymized microdata. Its main work may include: (1) discuss data privacy protection technologies, including user-oriented and data-oritented privacy protection, microdata publication, and k-anonymity; (2) define Privacy Inference Logic based on Set and Probability theories, and take it to formalize k-anonymity model; (3) detect and prevent potentially existed privacy inference attacks on the anonymized microdata, including simple and knowledge-based inference attacks; (4) propose several more reasonable data anonymization cost metrics (including knowledge-based cost metrics and the inference-proof cost metric) that can express the real information loss on the anonymized microdata in various situations, and mainly give two inference-proof policies: 1) considering inference attacks probabilities in data anonymization cost metric; 2) considering inference attacks before computing anonymization cost on each micarodata record.The main contributions include: (1) successfully detect several privacy inference attacks on anonymized microdata that have not been recognized in current literature, such as inference attacks under value domain, association, etc.; (2) propose Privacy Inference Logic and the formalized k-anonymity description, which is used for detecting and preventing privacy inference attacks on k-anonymized microdata; (3) the proposed data anonymization cost metrics and inference-proof policies are the core part for a real strong anti-inference k-anonymity model.Finally, the first-part experiment illustrates many kinds of inference attacks on anonymized microdata and their influence on privacy disclosure; the second-part experiment proves the efficiency and feasibility of the two inference-proof policies and the anonymization cost metrics in the paper.