Emotional Intelligence in Language Education: A Multimodal Analysis Framework Based on CLIP and Dynamic Weighting
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Within the field of language education—particularly in second language learning—emotional understanding between students and teachers is widely recognised as essential. It is crucial for enabling effective interaction and is closely associated with improved teaching and learning efficiency. This study presents a multimodal emotion analysis method that utilises a dynamic weighting mechanism, combined with a CLIP pre-trained model, to enhance emotional understanding. Using the Adaptive Language-Guided Multimodal Transformer (ALMT), the approach reconstructs cross-modal feature representations to boost the semantic quality of unimodal data, facilitating the integration of visual and linguistic features. The Adaptive Hyper Modal Learning Layer (AHL) features a dynamic weighting generation module that evaluates modality confidence in real-time, addressing information imbalances and reducing noise interference. Meanwhile, residual connections ensure stable parameter updates. Experimental results show a 3.2% increase in F1 score for emotion classification tasks on the CMU-MOSEI dataset, demonstrating improved robustness against missing modalities and noise. This method not only enhances the accuracy of emotion analysis but also supports emotional interaction in language education, thereby enriching the overall learning experience.
Fields of Study & Subjects
Emotion and Mood Recognition
Sentiment Analysis and Opinion Mining
Multimodal Machine Learning Applications
Publication details
- Journal / source
- IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
- Pages
- 735–742
- Publication date
- 28 January 2026
- Language
- EN