AG-MSA: Adaptive Gated Prompt Learning for Few-Shot Multimodal Sentiment Analysis

In the field of multimodal sentiment analysis (MSA), most existing approaches employ large pre-trained language models (PLMs) as the backbone for textual representation. However, these methods typically rely on full fine-tuning strategies, which not only incur substantial computational costs but also tend to suffer from overfitting in low resource or few-shot scenarios. To address these limitations, we propose a lightweight Adaptive Gated Prompt Learning framework for few-shot multimodal sentiment analysis. The proposed method dynamically assigns modality specific weights based on the semantic congruence between textual representations and the corresponding acoustic and visual signals, thereby effectively filtering out cross modal noise. Extensive experiments conducted on the CMU-MOSI and CMU-MOSEI benchmarks demonstrate that AG-MSA significantly outperforms full fine-tuning baselines while updating only a minimal parameter set, highlighting its effectiveness and robustness in low resource settings.