Supporting Relational Knowledge Construction Through LLM-Assisted Concept Mapping in Virtual Reality

Virtual reality (VR) is increasingly used to enhance learning, yet its effectiveness in supporting conceptual understanding, particularly relational knowledge construction, remains underexplored. Relational knowledge construction involves identifying meaningful connections among concepts to form coherent representations that support knowledge organization, guide interpretation, and facilitate learning transfer. Based on a review of the challenges in relational knowledge construction, we proposed conceptual, relational, and metacognitive scaffolds, upon which we developed CmapVR, an immersive learning environment with LLM-supported features to implement the proposed scaffolds. A within-subject study (N = 16) compared CmapVR with three alternative learning conditions in a science concept learning task. Results showed that CmapVR supported more extensive and higher-quality relational knowledge construction than the other conditions, resulted in lower extraneous cognitive load, and was preferred by participants over browsing in VR. We discussed factors that may have contributed to these outcomes and provided insights for integrating LLMs into VR-based learning environments.