CH-SIMS

Researchers propose SentiLLM, a framework that uses Semantic-Aligned Structural Abstraction to improve multimodal sentiment analysis by converting non-verbal signals into text-like tokens for large la…