Generating Privacy-preserving Educational Data Records with Diffusion Model

Educational Data Records (EDR) are crucial for capturing teaching behavior and student information, forming the basis for achieving educational intelligence. However, ensuring educational privacy has become a pressing concern, posing practical challenges to the use and sharing of educational data. To address the issue of EDR privacy preserving, we present EduSyn, a privacy data release scheme that utilizes generative diffusion models and differential privacy methods. Specifically, we adopt a diffusion modeling scheme that can be applied to both discrete and continuous types of data to accommodate the data characteristics of EDR, while an invariant Post Randomization (PRAM) perturbation method that satisfies local differential privacy is applied for data attributes that need to be specially protected before model training. We conduct comprehensive validation of this scheme within the domain of education applications, showcasing that EduSyn generates a superior private EDR dataset compared to similar generative methods and strikes a better privacy-utility trade-off.