Abstract
We present LLaVAC, a method for constructing a classifier for multimodal sentiment analysis. This method leverages fine-tuning of the Large Language and Vision Assistant (LLaVA) to predict sentiment labels across both image and text modalities. Our approach involves designing a structured prompt that incorporates both unimodal and multimodal labels to fine-tune LLaVA, enabling it to perform sentiment classification effectively. Experiments on the MVSA-Single dataset demonstrate that LLaVAC outperforms existing methods in multimodal sentiment analysis across three data processing procedures. The implementation of LLaVAC is publicly available at this https URL.
Abstract (translated)
我们介绍了LLaVAC方法,这是一种用于构建多模态情感分析分类器的方法。该方法利用大型语言和视觉助手(LLaVA)的微调来预测图像和文本两种模式的情感标签。我们的方法涉及设计一种结构化的提示,将单模态和多模态标签结合起来以对LLaVA进行微调,从而使它能够有效地执行情感分类任务。在MVSA-Single数据集上的实验表明,在三种不同的数据处理流程中,LLaVAC的性能超过了现有的多模态情感分析方法。LLaVAC的实现代码可在以下链接公开获取:[此URL](请将“此URL”替换为实际提供的GitHub或代码托管网站链接)。
URL
https://arxiv.org/abs/2502.02938