Target Hierarchy-Guided Knowledge Tracing: Fine-Grained Knowledge State Modeling
Xinjie Sun,Kai Zhang,Shuanghong Shen,Fei Wang,Yuxiang Guo,Qi Liu
DOI: https://doi.org/10.1016/j.eswa.2024.123898
IF: 8.5
2024-01-01
Expert Systems with Applications
Abstract:knowledge Tracing (KT) focuses on modeling the exercise process of students, assessing their knowledge state changes during the exercise process, and further providing targeted guidance for teaching and learning. Current KT models have made significant progress by utilizing deep learning technologies to model relevant attributes of the questions. However, in current knowledge tracing strategies, we have yet to exhaustively explore how to precisely evaluate the hierarchies of students’ knowledge mastery to better discern their current target hierarchy. To address the limitations of the KT model in accurately identifying and positioning target hierarchy, our research focuses on two key areas: first, we dynamically track students’ target hierarchies by evaluating their feedback on questions at different hierarchies ; second, we assess their actual question-solving abilities within these target hierarchies, thereby optimizing our evaluation of their knowledge status. Based on these insights, we propose the Target Hierarchy-guided Knowledge Tracing (THKT) model. In this model, we first incorporate the identification of question hierarchy into the model representation. Then, to avoid homogenization of the target hierarchy, we dynamically track the appropriate hierarchy for students based on their varying feedback to questions of the same hierarchy, thereby pinpointing their learning target hierarchies. Simultaneously, we develop a KCs applied ability learning module that works in conjunction with the target hierarchies to generate interpretable prediction results. The proposed THKT model has been tested and evaluated using three public, real-world educational datasets. The findings clearly demonstrate that our approach shines in the realm of KT prediction tasks, providing significantly interpretability. For broader research, we plan to provide the source code at: https://github.com/xinjiesun-ustc/THKT to ensure accessibility and promote further innovation in this field.