SemMSA Framework Uses LLM Semantics for Robust Multimodal Sentiment Analysis

SemMSA addresses incomplete multimodal data by using LLMs to construct latent sentiment semantics. The framework employs Cross-modal Semantic Refinement and anchor-free spectral alignment to integrate visual and acoustic representations without relying on noisy feature reconstruction.