Enhancing Multimodal Sentiment Analysis with Dynamic Weight Adjustment and Adversarial Training

Multimodal emotion recognition is a challenging task due to the complexity and variability of human emotions across different modalities. In this paper, we propose an innovative framework that integrates dynamic weight adjustment and adversarial training to enhance the robustness and accuracy of emotion recognition models. The dynamic weight adjustment module leverages statistical properties and entropy values to optimize modality weights, while the adversarial training module generates emotion-label-based adversarial samples to improve model resilience against noise and incomplete data. Extensive experiments on benchmark datasets demonstrate that our approach significantly outperforms existing methods in terms of both robustness and accuracy, making it a promising solution for real-world multimodal emotion recognition applications.