Multimodal Arabic Sentiment Analysis: A Comprehensive Review

Abstract

Multimodal Sentiment Analysis integrates textual, visual, and auditory information to achieve richer contextual understanding. Although multimodal learning has been advanced significantly in high-resource languages, research in Arabic remains limited and fragmented. This paper presents a comprehensive review of Multimodal Arabic Sentiment Analysis, to examine datasets, modeling paradigms, and fusion strategies. We survey available multimodal resources, analyze transformer-based architectures and large pretrained models across language, vision, and audio, and compare fusion mechanisms. Our review highlights persistent challenges, including limited dataset scale, dialectal diversity, modality imbalance favoring text, and trade-offs between modeling expressiveness and computational efficiency. We conclude by outlining key research gaps and future directions toward scalable, dialect-aware, and interaction-centric multimodal systems for the Arabic linguistic ecosystem.