Deep knowledge tracing with learning curves | JoVE Visualize

Area of Science:

Educational technology
Artificial intelligence in education
Machine learning for learning analytics

Background:

Knowledge tracing (KT) models predict student mastery using past responses, with deep learning models showing superior performance.
Existing deep KT models often overlook the learning curve theory, which posits that increased practice enhances concept mastery.
There is a need for KT models that explicitly incorporate learning curve dynamics to better reflect student learning progression.

Purpose of the Study:

To propose novel deep learning models, Convolution-Augmented Knowledge Tracing (CAKT) and Capsule-Enhanced CAKT (CECAKT), that explicitly model learning curves.
To enhance the prediction accuracy of student mastery by integrating the concept of learning curves into the KT framework.
To improve the ability of KT models to capture both overall knowledge state and concept-specific mastery.

Main Methods:

Developed CAKT using 3D convolutional neural networks to learn recent concept-specific experience, combined with LSTM for overall knowledge state.
Enhanced CAKT with CECAKT by replacing global average pooling with capsule networks to mitigate information loss.
Fused LSTM-derived overall knowledge state with convolutional or capsule-derived concept-specific features.

Main Results:

Both CAKT and CECAKT demonstrated superior performance compared to existing deep KT models across four real-life datasets.
The proposed models effectively integrated learning curve modeling into the knowledge tracing process.
Experimental results validated the enhanced predictive capabilities of CAKT and CECAKT.

Conclusions:

CAKT and CECAKT offer significant improvements over traditional deep KT models by incorporating learning curve theory.
These models provide a more nuanced understanding of student learning by considering both overall knowledge and concept-specific mastery progression.
The proposed approaches represent a promising advancement in intelligent tutoring systems and personalized learning platforms.