Deep learning based knowledge tracing in intelligent tutoring systems.

Area of Science:

• Educational Technology
• Artificial Intelligence
• Machine Learning

Background:

• Online education and intelligent tutoring systems (ITS) have grown, particularly during the COVID-19 pandemic.
• Knowledge tracing (KT) is crucial for ITS, modeling student knowledge states from interaction data to offer personalized feedback.
• Deep learning, like deep knowledge tracing, has advanced KT but often struggles with data sparsity.

Purpose of the Study:

• To propose a novel deep learning framework for knowledge tracing that effectively handles the challenge of data sparsity.
• To improve the accuracy of modeling and predicting student knowledge states in educational settings.

Main Methods:

• A quality-aware deep learning framework for knowledge tracing was developed.
• The framework incorporates sparse attention techniques and generative decoding to manage limited student interaction data.
• The proposed model was evaluated using extensive experiments on real-world datasets.

Main Results:

• The proposed quality-aware framework demonstrates accurate capture of student knowledge states.
• The method effectively addresses the data sparsity issue prevalent in existing knowledge tracing systems.
• Experimental results validate the framework's performance on diverse real datasets.

Conclusions:

• The developed quality-aware deep learning framework offers a robust solution for knowledge tracing with sparse data.
• This approach enhances the capability of intelligent tutoring systems to accurately assess and model student learning.
• The findings contribute to more effective and personalized online learning experiences.