Hierarchy-Aware Adaptive Forgetting with Differentiable Cognitive Stratification for Robust and Interpretable Knowledge Tracing

Knowledge Tracing (KT) is a core task for intelligent tutoring systems, which aims to model learners’ knowledge states and predict learning performance. To address the drawbacks of homogeneous forgetting decay and single-granularity cognitive modeling in existing KT models, this paper proposes the Hierarchy-Aware Adaptive Forgetting Knowledge Tracing (HAAF-KT) framework. Compared with existing models, the core innovation of this framework lies in the deep coupling of three-level cognitive hierarchy and adaptive forgetting mechanism for the first time. It adopts learnable hierarchical weighting to replace conventional hard discrete stratification, and introduces mask consistency regularization at the input level to boost noise robustness, achieving end-to-end joint optimization of cognitive stratification, personalized forgetting and anti-noise regularization. Extensive experiments on three public benchmark datasets including Algebra2005, ASSISTments2017 and ASSISTments2009 demonstrate that the proposed HAAF-KT achieves competitive prediction performance on all datasets. For Algebra2005, the AUC is 0.8728 and the ACC is 0.8154. For ASSISTments2017, the AUC is 0.8372 and the ACC is 0.7704. For ASSISTments2009, the AUC is 0.9314 and the ACC is 0.8536. All results are reported as the average of five independent repeated runs with standard deviations less than 0.001, which verifies the superior stability of the model. Meanwhile, the distribution entropy of the learned KCH hierarchy drops to 0.79, which is considerably lower than the entropy of uniform distribution (1.79). Validated by the levels-of-processing theory in cognitive psychology, the model can effectively mine discriminative cognitive hierarchical structures and support fine-grained and personalized knowledge tracing, and the learned cognitive representations carry explicit educational interpretability.

Authors

Institutions

Publication Details

Journal
Applied Sciences
Published
2026-11-16
DOI
https://doi.org/10.3390/app162211347
Primary Topic
Intelligent Tutoring Systems and Adaptive Learning
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Hierarchy-Aware Adaptive Forgetting with Differentiable Cognitive Stratification for Robust and Interpretable Knowledge Tracing

Fei Yu, Zhaorui Li, Jie Jin, Shanshan Xu
2 links
Applied Sciences
Intelligent Tutoring Systems and Adaptive Learning
article

Hierarchy-Aware Adaptive Forgetting with Differentiable Cognitive Stratification for Robust and Interpretable Knowledge Tracing

Fei Yu, Zhaorui Li, Jie Jin, Shanshan Xu
article en

Abstract

Knowledge Tracing (KT) is a core task for intelligent tutoring systems, which aims to model learners’ knowledge states and predict learning performance. To address the drawbacks of homogeneous forgetting decay and single-granularity cognitive modeling in existing KT models, this paper proposes the Hierarchy-Aware Adaptive Forgetting Knowledge Tracing (HAAF-KT) framework. Compared with existing models, the core innovation of this framework lies in the deep coupling of three-level cognitive hierarchy and adaptive forgetting mechanism for the first time. It adopts learnable hierarchical weighting to replace conventional hard discrete stratification, and introduces mask consistency regularization at the input level to boost noise robustness, achieving end-to-end joint optimization of cognitive stratification, personalized forgetting and anti-noise regularization. Extensive experiments on three public benchmark datasets including Algebra2005, ASSISTments2017 and ASSISTments2009 demonstrate that the proposed HAAF-KT achieves competitive prediction performance on all datasets. For Algebra2005, the AUC is 0.8728 and the ACC is 0.8154. For ASSISTments2017, the AUC is 0.8372 and the ACC is 0.7704. For ASSISTments2009, the AUC is 0.9314 and the ACC is 0.8536. All results are reported as the average of five independent repeated runs with standard deviations less than 0.001, which verifies the superior stability of the model. Meanwhile, the distribution entropy of the learned KCH hierarchy drops to 0.79, which is considerably lower than the entropy of uniform distribution (1.79). Validated by the levels-of-processing theory in cognitive psychology, the model can effectively mine discriminative cognitive hierarchical structures and support fine-grained and personalized knowledge tracing, and the learned cognitive representations carry explicit educational interpretability.

Applied SciencesVol. 16(22)
Hunan University of Science and Technology (CN), Changsha University of Science and Technology (CN)
National Natural Science Foundation of China
Quality education
Intelligent Tutoring Systems and Adaptive Learning
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon