A SURVEY OF TRANSFORMER-DRIVEN  HYBRID DEEP LEARNING METHODS FOR  MULTIMODAL SENTIMENT ANALYSIS IN  NLP

Anitha P; Sivagami R

Paper Title

A SURVEY OF TRANSFORMER-DRIVEN HYBRID DEEP LEARNING METHODS FOR MULTIMODAL SENTIMENT ANALYSIS IN NLP

Authors

Anitha P

Sivagami R

Keywords

Unimodal sentiment analysis
Multimodal sentiment analysis
fusion methodologies
textual
visual
and audio/visual data

Journal

International Journal of Research and Analytical Reviews

Publication Info

Volume: 12 | Issue: 4 | Pages: 37-47

Published On

September, 2025

Downloads

FULL PDF

CITATION

COPY LINK

Abstract

Sentiment analysis (SA) is a computational approach aimed at identifying, extracting, and quantifying subjective information such as emotions, attitudes, and opinions. While traditional unimodal sentiment analysis relies solely on textual data, multimodal sentiment analysis (MSA) leverages diverse information sources, including speech, tone, facial expressions, and body movements, to capture a deeper, more accurate representation of human emotions. This study provides a comprehensive review of existing research on multimodal fusion methods and feature extraction, emphasizing the integration of textual, visual, and audio-visual data through transformer-based deep learning models. It traces the evolution and theoretical underpinnings of MSA, outlining its major advancements, existing challenges, and practical advantages. Furthermore, the paper identifies emerging trends and future research directions, serving as a valuable reference for scholars and practitioners seeking to advance the field of multimodal sentiment analysis.