Scene-aware Adaptive Multimodal Fusion Based Student Learning Emotion Identification

Empirical evidence demonstrates that students' emotional states significantly influence their academic performance; however, emotions in the classroom are characterized by their subtle contextual relevance. Emotional cues are distributed across various sensory modalities and undergo dynamic changes during different teaching and learning activities, making accurate emotion recognition a challenging task. In this paper, we propose a Scene-Aware Adaptive Multimodal Fusion (SAMF) framework along with an Adaptive Modal Scoring (AMS) mechanism that selectively integrates the most relevant combination of modalities based on scene labels—such as lecture, discussion, or Q&A session. This adaptive process enhances the SAMF framework's ability to detect subtle and contextually relevant emotional cues, thereby improving the accuracy of emotion recognition in dynamic educational environments. Experimental results indicate that the algorithm significantly outperforms traditional methods in complex classroom settings, demonstrating that the SAMF framework delivers timely, sensitive, and accurate emotion recognition, which offers robust technical support for real-time monitoring of classroom activities and dynamic interventions.