Improving Question Embeddings With Cognitive Representation Optimization for Knowledge Tracing.

Area of Science:

• Educational Technology
• Artificial Intelligence in Education
• Cognitive Science

Background:

• Existing knowledge tracing (KT) models predict student performance using historical records but often ignore distractors like slipping and guessing.
• Static cognitive representations in current KT models are limited and may not accurately reflect dynamic student understanding.
• This can lead to issues in the synergy and coordination of student learning data.

Purpose of the Study:

• To propose a novel Cognitive Representation Optimization for Knowledge Tracing (CRO-KT) model.
• To enhance the accuracy of predicting student knowledge status and future performance.
• To address limitations of static cognitive representations and incorporate the impact of distractors.

Main Methods:

• Utilizes a dynamic programming algorithm to optimize the structure of cognitive representations based on exercise difficulty.
• Employs a co-optimization algorithm to refine cognitive representations by considering co-related exercises.
• Fuses learned relational embeddings from bipartite graphs with optimized record representations for richer cognitive expression.

Main Results:

• The CRO-KT model demonstrates effectiveness in optimizing cognitive representations for knowledge tracing.
• Experimental validation on three public datasets confirms the model's superior performance.
• The approach enhances the expression of student cognition by integrating various optimization techniques.

Conclusions:

• The proposed CRO-KT model offers a significant advancement in knowledge tracing.
• Optimizing cognitive representations and accounting for distractors leads to more accurate student performance predictions.
• This research provides a more robust framework for understanding and modeling student learning processes.