Transformer-based convolutional forgetting knowledge tracking.

Area of Science:

• Educational Technology
• Artificial Intelligence
• Machine Learning

Background:

• Knowledge tracking is crucial for personalized online learning, analyzing student mastery via learning history.
• Traditional models like CNNs struggle with long-term dependencies, while Transformers offer improvements but can overlook connections in repetitive training data.
• Addressing limitations in existing Transformer models for knowledge tracking is essential for accurate learning path recommendations.

Purpose of the Study:

• To develop an advanced knowledge tracking model that overcomes the limitations of standard Transformer architectures.
• To enhance the accuracy of knowledge tracking by incorporating contextual information and simulating the student forgetting phenomenon.
• To introduce the Transformer-based Convolutional Forgetting Knowledge Tracking (TCFKT) model.

Main Methods:

• Introduced a convolutional attention mechanism to improve the model's perception of contextual information.
• Simulated the student forgetting phenomenon using a forgetting factor, integrated with the model's weight matrix.
• Developed and evaluated the Transformer-based Convolutional Forgetting Knowledge Tracking (TCFKT) model.

Main Results:

• The TCFKT model demonstrated superior performance compared to existing knowledge tracking models.
• Experimental results on real-world datasets (ASSISTments2012, ASSISTments2017, KDD a, STATIC) validated the model's effectiveness.
• The proposed model successfully addressed issues of ignored connections in repetitive training data.

Conclusions:

• The TCFKT model represents a significant advancement in knowledge tracking for online education.
• The integration of convolutional attention and forgetting factors enhances the accuracy and robustness of knowledge tracking.
• TCFKT offers a more effective approach to analyzing student knowledge mastery and informing personalized learning recommendations.