Answers are grounded in this paper and cite their sources.
Human learning is a dynamic cognitive process in which latent knowledge and competency states evolve through sequential interactions. However, existing generative AI learning-support systems often separate learner-state modeling, resource recommendation, knowledge-grounded generation, and pedagogical planning, limiting adaptive performance and interpretability. This study proposes ICG-AGLS, an interdisciplinary competency graph-guided adaptive generative AI learning support framework. It models learner states from historical interactions, projects them into a competency space, and constructs an interdisciplinary competency graph linking courses, concepts, resources, assignments, and competency dimensions. The framework integrates competency-gap-guided recommendation, graph-guided retrieval-augmented generation, adaptive reasoning routing, and multi-agent pedagogical coordination. Experiments on MOOCCubeX, OULAD, and EdNet demonstrate consistent improvements across learner-state modeling, recommendation, knowledge-grounded generation, and multi-agent planning. ICG-AGLS achieves an AUC of 0.871, Recall@10 of 0.812, ROUGE-L of 0.772, and a task completion rate of 0.839, outperforming representative task-specific baselines. These findings suggest that competency-structured latent-state modeling provides an interpretable computational bridge between evolving human learning states and adaptive generative AI at the behavioral and computational-representation levels, rather than establishing direct neural correspondence. The framework offers practical potential for personalized learning support in higher education.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: