Knowledge Tracing via Attention Enhanced Encoder-Decoder.

Area of Science:

• Educational Technology
• Artificial Intelligence
• Machine Learning

Background:

• Traditional knowledge tracing models use limited learning data, failing to capture complex student behaviors.
• Existing models struggle to integrate different learning behavior types, hindering accurate collaboration analysis.
• Accurate student knowledge state assessment is crucial for personalized learning and effective educational interventions.

Purpose of the Study:

• To develop an improved knowledge tracing model that synthesizes multiple learning behavior types.
• To enhance the modeling of constraints within learning behaviors and collaboration across behaviors.
• To achieve superior performance in predicting student knowledge states and future answers.

Main Methods:

• Proposed a novel knowledge tracing model utilizing an attention-enhanced encoder-decoder architecture.
• Implemented multi-headed attention mechanisms to capture intra-behavior constraints.
• Employed channel attention mechanisms to model inter-behavior collaboration.

Main Results:

• The proposed model demonstrated superior performance compared to existing knowledge tracing methods on real-world datasets.
• Experimental results indicated enhanced accuracy in knowledge state representation.
• The model achieved better predictive performance in practical applications on an intelligent learning platform.

Conclusions:

• The attention-enhanced knowledge tracing model effectively integrates diverse learning behaviors for improved accuracy.
• The model's ability to capture constraints and collaboration enhances its utility in intelligent tutoring systems.
• This approach offers a significant advancement in personalized education and learning analytics.