Too Detailed to Share? Towards Risk-Based Privacy Protection of Fine-Grained Educational Data

Abstract

Due to the recent increase in the use of digital learning platforms, fine-grained digital-trace data has been growing in the education sector. However, despite the potential of such micro-level log data for understanding and personalising individual learning processes, its secondary use is limited due to privacy concerns. A key to advancing data sharing for the secondary use while protecting individual privacy is effective risk assessments. Nevertheless, prior research predominantly focuses on privacy risks of structured tabular data, leaving fine-grained digital-trace data underexplored. To fill this gap, we conduct a comprehensive risk analysis of fine-grained educational data using the unicity framework. Employing two real-world datasets reflecting on secondary and higher education settings and two open datasets on self-paced language learning, we demonstrate that fine-grained educational data is highly susceptible to re-identification through timestamps. In addition, we show that the effectiveness of naive coercing of timestamps depends on the number of students in the dataset and the diversity of educational contexts where the data is collected. Our findings help practitioners to make risk-based decisions to choose appropriate privacy protection strategies.