An optimal approach for predicting cognitive performance in education based on deep learning

Abstract

Research background

Knowledge tracking in educational data mining has become increasingly important for identifying students' knowledge gaps and enhancing individualized instruction. Traditional exercise recommendation algorithms often overlook students' forgetting behaviour, hindering effective learning.

Purpose

This study aims to develop a novel approach to integrating the law of forgetting into a deep learning-based knowledge-tracking model, improving exercise recommendations and effectively addressing students' learning gaps.

Methods

The proposed knowledge probability prediction model incorporates the forgetting curve theory alongside a dynamic key-value memory mechanism for tracking students' knowledge mastery levels. This model continuously adapts to students' interactions, allowing personalized exercise recommendations considering mastered and forgotten knowledge points.

Results

The proposed model was evaluated using ASSISTment 2009, Statics 2011, and DouDouYun datasets. The results indicate that our approach significantly outperforms traditional recommendation algorithms regarding novelty and concept coverage, effectively accommodating students' forgetting behaviour.

Conclusion

Integrating the forgetting law into knowledge-tracking systems leads to more effective and personalized exercise recommendations, ultimately facilitating improved learning outcomes. This approach enhances students' acquisition of new knowledge and more efficiently addresses their existing knowledge gaps.