Multimodal Affective Sensing Driven Adaptive Intervention Model for Second Language Learning Stress

In high-pressure learning contexts, second language (L2) learners often struggle with the interplay between anxiety and cognitive overload, significantly impairing their language acquisition efficiency. However, most existing intelligent education systems fail to dynamically detect and regulate learners psychological states throughout the learning process. To address this gap, this study proposes an adaptive intervention model that integrates multimodal emotion recognition with reinforcement learning-based strategy optimization. By capturing learners facial expressions, vocal prosody, and physiological signals at key task stages, the system enables real-time emotional state recognition and generates personalized intervention strategies based on emotional feedback. Sixty university-level English learners were recruited for an 8-week randomized controlled trial. While the control group followed a conventional instructional approach, the experimental group was supported by the emotion-sensing intervention system. The study employed standardized language tests and system performance metrics to assess the effectiveness of the intervention, supplemented by learning logs and interviews to collect subjective user feedback. Results show that the experimental group outperformed the control group in terms of emotion recognition accuracy, intervention response latency, and language performance improvement. Learners also reported high acceptance and positive evaluations of the system. This research validates the feasibility of multimodal affective sensing in mitigating learning stress and provides both technical and empirical foundations for emotion-adaptive intelligent education systems, thereby expanding the application boundary of affective computing in intelligent instruction.