Enhancing Privacy in Lightweight Data Encoding for Sensitive Applications

Record linkage aims to identify records referring to the same individual across different datasets, but often involves sensitive personal identifiers. Privacy-Preserving Record Linkage (PPRL) techniques, particularly those based on Bloom filter encoding, enable secure linkage while protecting privacy. However, existing methods suffer from information loss and lack rigorous privacy guarantees. To address these limitations, we propose a novel (Hash)-A method that retains position and frequency information of q-grams, enhancing matching accuracy. We also introduce the Utility-optimized Bloom Filter (UBF), a privacy-preserving mechanism grounded in User-Level Local Differential Privacy (ULDP), which offers formal and quantifiable privacy guarantees. These components are integrated into a complete three-party PPRL protocol that supports missing value handling. Experimental results on real-world datasets show that our method achieves improved linkage accuracy and robust privacy protection against inference attacks, demonstrating a practical balance between utility and privacy.