LLM-Based System for Multimodal Sentiment Analysis

This paper addresses the challenging task of multimodal conversational aspect-based sentiment analysis, a crucial area for building emotionally intelligent AI. It focuses on two subtasks: extracting a sentiment sextuple and detecting sentiment flipping. The use of structured prompting and LLM ensembling demonstrates a practical approach to improving performance on these complex tasks. The results, while not explicitly stated as state-of-the-art, show the effectiveness of the proposed methods.

• •Employs structured prompting to guide LLMs for sentiment component extraction.
• •Utilizes LLM ensembling to improve sentiment flipping detection.
• •Demonstrates effectiveness in a multimodal conversational aspect-based sentiment analysis challenge.