GELT: A graph embeddings based lite-transformer for knowledge tracing.

Area of Science:

• Artificial Intelligence in Education
• Machine Learning for Learning Analytics

Background:

• Knowledge tracing is crucial for intelligent education, traditionally relying on student performance assessment.
• Deep Learning models offer complex representations but often lack interpretability due to end-to-end training.
• Existing methods struggle with understanding skill-question relationships and computational efficiency.

Purpose of the Study:

• To develop an interpretable and computationally efficient model for knowledge tracing.
• To uncover and understand the relationships between educational skills and assessment questions.
• To improve the prediction accuracy of student knowledge states.

Main Methods:

• Proposed GELT (Graph Embeddings based Lite-Transformer), inspired by graph neural networks.
• Introduced an energy-saving attention mechanism for knowledge state prediction.
• Utilized three publicly available real-world datasets for knowledge tracking.

Main Results:

• GELT demonstrates superior performance compared to state-of-the-art baselines.
• The model effectively uncovers skill-question relationships.
• The energy-saving attention mechanism significantly reduces computational costs while maintaining high accuracy.

Conclusions:

• GELT offers an interpretable and efficient solution for knowledge tracing in intelligent education.
• The proposed approach advances the field of learning analytics by integrating graph embeddings and efficient attention mechanisms.
• This work provides a foundation for more transparent and resource-aware educational AI systems.