A Comprehensive Review of Multimodal Sentiment Analysis in NLP using Deep Learning

Abstract

Sentiment analysis, whether executed within a unimodal or multimodal paradigm, is commonly referred to as opinion mining. It represents a computational methodology used to identify, extract, and quantify subjective information, including perspectives, attitudes, and emotional states. Unlike traditional unimodal sentiment analysis that relies solely on text, multimodal sentiment analysis (MSA) incorporates information from multiple sources—such as speech, tone of voice, facial expressions, and body gestures—to provide a richer and more accurate understanding of emotions. This study presents an extensive survey of research on multimodal fusion techniques and features, with a particular focus on the integration of textual, visual, and audio-visual data. This scholarly investigation explores the historical evolution and theoretical foundations of Multimodal Sentiment Analysis (MSA), examining both current challenges and its benefits. Moreover, this manuscript highlights potential directions for future research, making it a valuable resource for both academic and industry researchers in this domain.