Hybrid framework of differential privacy and secure multi-party computation for privacy-preserving entity resolution

The exponential improvement and precision in hardware design, coupled with sophisticated software systems, are the basis of unprecedented rates of data generation and storage. However, extracting actionable knowledge, formulating impactful policies, and making insightful decisions from these massive datasets rely on data integration with entity resolution as its core task. Despite significant advances in entity resolution methods, the risk of data breaches, matching accuracy, utility and scalability remain critical challenges to the data science research community. This study introduces a novel hybrid framework of differential privacy (DP) and secure multi-party computation (SMPC) for privacy-preserving entity resolution (PPER), thereby addressing critical data utility and confidentiality challenges. We rigorously evaluated the framework using the Febrl4 and North Carolina Voter Registration (NCVR) datasets across three supervised machine learning models (Logistic Regression, SVM, Naïve Bayes), through adaptive ε-allocation (0.1 to 5.0), demonstrating the crucial privacy-utility trade-off. Our findings reveal that the framework maintains high linkage utility, with F1-scores consistently above 0.81 even under stringent privacy budgets (ϵ=0.1), and achieving over 0.90 at moderate ϵ values, notably with support vector machine exhibiting robust performance. This research provides empirical evidence and theoretical guarantees for developing highly practical and ethically compliant PPER solutions, offering clear guidance for balancing data utility with privacy requirements across diverse application domains.