DKVMN&MRI: A new deep knowledge tracing model based on DKVMN incorporating multi-relational information.

Area of Science:

• Educational Technology
• Artificial Intelligence
• Cognitive Science

Background:

• Knowledge tracing is crucial for intelligent education systems, modeling student knowledge states from historical data.
• Existing models struggle with sparse data, interpretability, and capturing complex exercise relationships.

Purpose of the Study:

• To develop an advanced deep knowledge tracing model (DKVMN&MRI) addressing limitations of current approaches.
• To enhance the prediction of student learning ability by integrating multiple relational data types.

Main Methods:

• Utilized Dynamic Key-Value Memory Network (DKVMN) with Long Short-Term Memory (LSTM) for modeling learning processes.
• Incorporated the Ebbinghaus forgetting curve to simulate memory decay.
• Integrated Item Response Theory (IRT) and attention mechanisms for predictive accuracy.

Main Results:

• DKVMN&MRI demonstrated significant improvements in AUC and ACC metrics across three real-world datasets.
• The model effectively captures exercise-knowledge point, exercise-exercise, and learning-forgetting relationships.
• Achieved superior performance compared to state-of-the-art knowledge tracing models.

Conclusions:

• DKVMN&MRI offers a more accurate and interpretable approach to knowledge tracing.
• The model's ability to integrate diverse relationships enhances its efficacy in intelligent education.
• Provides valuable insights into learner knowledge states and exercise interactions.