A Knowledge Query Network Model Based on Rasch Model Embedding for Personalized Online Learning.

Area of Science:

• Education Technology
• Data Science
• Machine Learning

Background:

• Online education generates vast amounts of data, necessitating advanced analysis techniques.
• Knowledge tracing models learner progress using exercise data but often lacks personalization and interpretability.
• Existing deep learning models overlook personalized question and learner information, limiting insights.

Purpose of the Study:

• To propose a Context-Aware Attentive Knowledge Query Network (CAKQN) model for enhanced knowledge tracing.
• To integrate personalized representations and improve interpretability in educational data analysis.
• To effectively model learner sequences, contextual information, and forgetting behavior.

Main Methods:

• Developed the Context-Aware Attentive Knowledge Query Network (CAKQN) model.
• Utilized the Rasch model for regularizing learner-question interactions and obtaining personalized embeddings.
• Employed Long-Term Short-Term Memory (LSTM) networks and a monotonic attention mechanism to analyze sequential data and learner behavior, including forgetting.

Main Results:

• CAKQN demonstrated superior performance on four real-world datasets, improving AUC by an average of 2.945% over existing optimal models.
• The model successfully tracked learner knowledge status and provided enhanced interpretability.
• CAKQN effectively captured learner forgetting behavior, a characteristic often missed by other models.

Conclusions:

• The CAKQN model offers a significant advancement in knowledge tracing for online education.
• The model's ability to personalize and interpret learner interactions holds promise for intelligent tutoring systems.
• Future applications include personalized learning strategies, adaptive teaching interventions, and optimized resource recommendations.