Multi-level visual-textual alignment transformer for multimodal aspect-based sentiment analysis

Hongxin Li

doi:10.1016/j.eswa.2025.130148

Multi-level visual-textual alignment transformer for multimodal aspect-based sentiment analysis

2025-10-25

0

PRE

AI

DOI：10.1016/j.eswa.2025.130148

Original

Original request for help

Save

Abstract

En 中文

• We propose an MVAT approach for multimodal aspect-based sentiment analysis. • Align image and text by minimizing the Wasserstein distance of tokens and patches. • Designed a dynamic multimodal gate mechanism to integrate features from all directions. • We designed a GCN-based Differential Transformer to enhance focus on target aspect terms.

Multi-level visual-textual alignment transformer for multimodal aspect-based sentiment analysis

Abstract

Journal

Organization