A Multimodal Neuro‐Adaptive Tutoring Agent for Children's Classical Chinese Poetry Learning in VR

ABSTRACT

This work offers a dual‐loop neural‐adaptive VR teaching agent to mitigate cognitive load fluctuations, enhance abstract semantic comprehension, and counteract declining motivation in children's classical Chinese poetry learning. The system amalgamates multimodal inputs, such as EEG and eye tracking, to incessantly deduce learners' cognitive states and synchronize instructional feedback across micro‐ and macro‐level loops. An inference engine based on random forests was developed, employing a heuristic pseudo‐labeling method to address label scarcity. Offline assessments validated consistent classification across various cognitive states. In the micro loop, inferred states prompt immediate task‐specific interventions; in the macro loop, they govern the rate of learning and the timing of reviews, establishing a dual‐loop system that connects momentary learning with long‐term memory organization. A controlled experiment with a between‐subjects design assessed system‐level impacts in genuine poetry learning contexts. In comparison to a nonadaptive VR baseline, the neural‐adaptive system markedly improved children's comprehension of poetic imagery and semantics, augmented memory retention, diminished perceived cognitive load, and heightened learning desire. The findings indicate that integrating real‐time cognitive state awareness with dual‐loop adaptive training provides an efficient and pragmatic framework for immersive classical poetry education in youngsters.