Abstract As the scale of online learning expanded, students tended to exhibit declining learning behaviors and accumulating task backlogs during self-directed study. These issues further led to reduced learning motivation and increased psychological pressure. To address this, this study constructed an innovative model that integrated a knowledge graph with multi-agent reinforcement learning. The model enabled learning state risk identification and personalized intervention. The study conducted a systematic evaluation using comprehensive learning behavior data from seven selected courses. The results indicated that the model achieved a high level of risk identification at an early stage. The recall values for all courses ranged from 0.879 to 0.896. As the learning process progressed, accuracy steadily increased to above 0.889. The F1-score remained between 0.842 and 0.871 across all stages, which demonstrated strong stability. Furthermore, the intervention strategies significantly improved learning trajectories across two experimental semesters. Students’ learning activities showed continuous improvement over time. Behavioral fluctuations and breakpoint frequency were both markedly reduced. These findings confirmed that the model consistently enhanced learning motivation, stabilized learning rhythms, and optimized patterns of resource utilization.
Guo et al. (Fri,) studied this question.