Multimodal LLMs See Sentiment 事件

Multimodal LLMs See Sentiment arXiv:2508.16873v3 Announce Type: replace Abstract: Understanding how visual content conveys sentiment is increasingly important in a digital landscape dominated by imagery. However, sentiment perception depends on complex scene-level semantics, making this a challenging task for computational models. This paper examines how Multimodal Large Language Models (MLLMs) perform sentiment analysis in images through a systematic, evaluation-driven study encompassing three