η-Inference: A Data-Aware and High-Utility Privacy Model for Relational Data Publishing

Current privacy-preservation data publishing technologies primarily focus on anonymizing datasets, often overlooking the inherent privacy degrees embedded within the data. This oversight makes it challenging to balance privacy and utility effectively. To address this issue, we introduce the Privacy Evaluation and Validation scheme to Measure the original Privacy Degree (PEVMPD), anchored in a novel η-inference model. PEVMPD operates in two phases: the identification of risk elements and the evaluation of privacy degrees. In the first phase, attributes are appraised using entropy and KL divergence to pinpoint sensitive attributes. Concurrently, the maximum entropy principle is employed to identify critical quasi-identifiers. The second phase involves applying these risk elements within our η-inference model to locate data vulnerable to privacy breaches and to quantify the corresponding privacy degree. This methodology enables data owners to make informed decisions about achieving an optimal privacy-utility tradeoff during data anonymization. Experimental results on real datasets validate the effectiveness of PEVMPD in enhancing privacy measures.