A Generative AI-Driven Multimodal Assessment and Intervention System for Metacognition in Online Learning

Metacognition constitutes a critical determinant of self-regulated online learning, and its effective cultivation represents a significant focus in online education research. Existing interventions predominantly emphasize academic performance diagnostics, lacking targeted strategies for intrinsic competencies and systematic multimodal data integration, which constrains both comprehensiveness and accuracy. This study proposes a generative AI-driven multimodal assessment and intervention system for metacognition in online learning. The framework employs self-attention Bi-LSTM, VGG-16, Wav2vec 2.0+BERT, and BP neural networks to evaluate metacognitive dimensions across textual, visual, acoustic, and behavioral modalities. Decision-level multimodal fusion computes individualized weightings, which subsequently inform a generative AI to produce tailored intervention protocols. Deployment across multiple online courses at South China Normal University demonstrated that the multimodal architecture achieved precise evaluation and significantly enhanced learners' metacognitive competencies.