Related Experiment Videos
SRP-Net: sensitive risk propagation network with asymmetric cross-attention for educational data.
Junpeng Hu1,2, Xiao Guo1, Tao Hu1
1College of Intelligent Systems Science and Engineering , Hubei Minzu University, Enshi, 445000, China.
Scientific Reports
|June 4, 2026
Summary
We developed Sensitive Risk Propagation Network (SRP-Net) to detect sensitive educational data, overcoming semantic mismatches and improving risk assessment accuracy. This novel approach enhances data privacy in digital education.
Area of Science:
- Computer Science
- Data Science
- Educational Technology
Background:
- Digital transformation in education generates vast structured data, posing challenges for sensitive information detection.
- Existing methods struggle with semantic mismatches between data fields and isolated risk assessments, ignoring inter-field relevance.
- The need for robust privacy-preserving techniques in educational data is critical.
Purpose of the Study:
- To propose a novel deep learning architecture, Sensitive Risk Propagation Network (SRP-Net), for accurate sensitive information detection in structured educational data.
- To address the limitations of semantic mismatch and isolated risk assessment in current approaches.
- To introduce a new benchmark dataset and domain-specific language model for advancing research in educational data privacy.
Main Methods:
- Developed a sequential two-stage architecture (SRP-Net) employing asymmetric cross-attention for in-field alignment and a sensitivity-aware graph neural network for cross-field propagation.
- Utilized batch dynamic subgraph extraction within the graph neural network to prevent information smoothing.
- Introduced Edu KG BERT for domain adaptation via structured knowledge injection and created the Edu-Sens dataset using privacy-preserving protocols.
Main Results:
- SRP-Net achieved a Macro-F1 score of 0.8779 and 95.58% accuracy on the Edu-Sens benchmark dataset.
- Demonstrated a 3.06% improvement over non-graph baselines and significantly outperformed traditional machine learning and pure semantic methods.
- Ablation studies validated the effectiveness of both in-field alignment and dynamic cross-field propagation.
Conclusions:
- SRP-Net effectively detects sensitive information in structured educational data by integrating in-field alignment and cross-field propagation.
- The proposed methods and resources (Edu KG BERT, Edu-Sens) advance the field of educational data privacy and security.
- The findings highlight the importance of holistic risk assessment that considers field relevance and inter-field relationships.
