A knowledge tracing approach with dual graph convolutional networks and positive/negative feature enhancement network

Area of Science:

• Educational Data Mining
• Artificial Intelligence in Education
• Machine Learning

Background:

• Knowledge tracing models are crucial for predicting student mastery of skills.
• Existing models face challenges with scalability, data sparsity, individual learning differences, and skill interdependencies.

Purpose of the Study:

• To develop an advanced knowledge tracing method addressing limitations of current approaches.
• To enhance the accuracy and robustness of student mastery prediction in complex educational settings.

Main Methods:

• Proposed a novel knowledge tracing method utilizing dual graph convolutional networks (GCNs) and positive/negative feature enhancement.
• Constructed dual graph structures representing students and skills, processed independently by GCNs.
• Integrated feature enhancement and spectral embedding clustering for efficient feature combination and optimization.

Main Results:

• The proposed dual GCN method significantly outperformed existing knowledge tracing approaches on public datasets.
• Demonstrated superior performance in handling data sparsity and complex skill correlations.
• Effectively addressed variations in individual learning performance.

Conclusions:

• The developed method offers a promising advancement for educational data mining and personalized learning.
• Highlights the potential of graph learning models and feature enhancement techniques in educational applications.