Fine-grained sentiment analysis based on cross-modal information translation

期刊论文
发表日期：

2025年5月8日

查看全文

Shaowu Zhang,
Pengyuan Du,
Xijun Cui,
Hongfei Lin,
Liang Yang

摘要

Fine-grained sentiment analysis has been a hot topic in the field of natural language processing in recent years, and the diverse ways people express sentiments necessitate multi-modal sentiment analysis with text modality enhancement. In this paper, we propose a modal information translation approach to address the challenges arising from information alignment across different modalities, such as discrepancy between image and text information, and we use a graphical textual model for modal enhancement. At the same time, using text information and image information, fusing multi-source information and effectively integrating them into a unified representation enables the model to capture the relationship between text semantics and image entities more comprehensively, thus improving the expressive power and performance of the model. The F1 value of the model for the sentiment categorization task on the Twitter15 and Twitter17 datasets is 72.9 $\%$ $%$ and 70.6 $\%$ $%$ , respectively. The experimental results demonstrate that the modal information translation method based on graph-generated text performs better on the fine-grained sentiment analysis task.