Adaptive Knowledge Tracing with Dynamic Memory and Reinforcement Learning | JoVE Visualize

Area of Science:

Educational Technology
Artificial Intelligence in Education
Cognitive Science

Background:

Conventional knowledge tracing (KT) methods struggle with data sparsity, learner forgetting, and static cognitive models, hindering personalized learning.
Accurate assessment of student knowledge states and dynamic adaptation are crucial for effective personalized learning.
Existing KT models often fail to capture the dynamic nature of cognitive changes and forgetting processes.

Purpose of the Study:

To propose Dynamic Reinforcement learning-based Adaptive Knowledge Tracing (DRAKT), a novel model designed to overcome limitations of traditional KT approaches.
To enhance the robustness and accuracy of knowledge tracing by incorporating dynamic updates and forgetting mechanisms.
To provide a reliable technical foundation for personalized learning-path recommendation and real-time cognitive adaptation in intelligent educational systems.

Main Methods:

Developed DRAKT, integrating a Q-learning-based mechanism with the Ebbinghaus forgetting curve for dynamic knowledge-state adjustment.
Implemented a dynamic memory update module using a gated recurrent unit (GRU) and attention-based filtering to capture learning dependencies.
Validated DRAKT on three public ASSISTments datasets (2009, 2012, 2017) against state-of-the-art baselines.

Main Results:

DRAKT consistently outperformed existing state-of-the-art knowledge tracing models across multiple datasets.
Achieved high AUC scores (82.08% on ASSISTments2017, 81.47% on ASSISTments2009), significantly surpassing the next best model (GKT).
Demonstrated substantial improvements in accuracy, outperforming GKT by 4.77-5.75 percentage points.

Conclusions:

DRAKT effectively addresses data sparsity, forgetting, and dynamic cognitive changes in knowledge tracing.
The proposed model offers a significant advancement in accurately assessing and adapting to individual student learning processes.
DRAKT provides a robust framework for developing more effective intelligent educational systems for personalized learning.