Personalized Forgetting-aware Multi-view Graph Knowledge Tracing for Vocational Bachelor’s Practical Training Platform Evaluating

Abstract

Vocational bachelor’s practical training platforms play an important role in cultivating high-level technical talents, where accurately evaluating students’ learning progress is essential for personalized guidance and adaptive training optimization. Knowledge tracing provides an effective way to infer students’ evolving knowledge states from historical learning interactions. However, existing methods still suffer from two limitations: they usually fail to jointly model static knowledge structures and dynamic cognitive evolution, and they often ignore personalized forgetting behaviors caused by time intervals, skill characteristics, and repeated practice. To address these issues, this paper proposes a personalized forgetting-aware multi-view graph knowledge tracing method for student performance evaluation in vocational bachelor’s practical training platforms. Specifically, a multi-view graph aggregation module is first constructed, where a heterogeneous graph captures high-order semantic relations among students, exercises, and skills, while a session graph models the temporal evolution of students’ response-aware knowledge states. Then, a personalized forgetting-aware calibration module is designed to estimate individualized forgetting intensity by jointly considering hidden knowledge states, exercise representations, skill representations, time intervals, and historical practice frequency. Based on the calibrated knowledge state, an adaptive performance evaluation module integrates target exercise information, related skill representations, and historical learning experience for response prediction. Furthermore, a forgetting-aware adaptive loss is introduced to emphasize cognitively unstable and uncertain interactions during optimization. Experimental results on benchmark datasets show that the proposed method consistently outperforms representative knowledge tracing baselines in terms of AUC, ACC, and Recall.