Dynamic domain learning ability enhanced knowledge tracing with stability.

Area of Science:

• Educational Technology
• Artificial Intelligence in Education
• Cognitive Science

Background:

• Existing Knowledge Tracing (KT) models struggle with temporal fluctuations and static ability modeling.
• Current methods overlook the gradual, stable nature of knowledge evolution.
• Individual differences in task-specific abilities are not adequately captured.

Purpose of the Study:

• To develop a novel Knowledge Tracing (KT) model that addresses limitations in temporal stability and personalized ability assessment.
• To enhance the accuracy and stability of predicting student learning performance.
• To incorporate domain learning ability and knowledge dynamics into the KT framework.

Main Methods:

• Proposed Dynamic Domain Learning Ability Enhanced Knowledge Tracing with Stability (DLAKT) model.
• Explicitly incorporated domain learning ability, mapping skills to multiple ability dimensions.
• Utilized memory networks and a Transformer-based smoothing module to model knowledge dynamics and stability.
• Dynamically adjusted ability improvement rates based on knowledge state, response time, and item difficulty.

Main Results:

• DLAKT demonstrated superior prediction accuracy compared to existing mainstream models across three real-world datasets.
• The model successfully captured gradual knowledge evolution and reduced temporal fluctuations in predictions.
• Personalized modeling of evolving student abilities was achieved through dynamic updates.

Conclusions:

• DLAKT provides a more accurate and stable approach to Knowledge Tracing by integrating domain learning ability and advanced modeling techniques.
• The model's interpretability is enhanced through explicit representation of domain learning ability.
• This research advances the field of AI in education by offering a more robust tool for understanding and predicting student learning trajectories.