Multimodal Emotion Recognition using Deep Feature Fusion and Attention Mechanism: A Survey

Read full text Summarise with VIRAT
Authors: Faisal Majeed, Poonam Dhankhar
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Multimodal Emotion Recognition (MER) is a central area of affective computing offering a more precise, accurate and robust understanding of human emotional states. It is achieved by the integration of multiple data streams at a time like text data, speech data, and visual data information.1 This report offers a comprehensive survey of the methodologies that are driving the modern MER systems with a main focus on the integration of deep feature fusion and advanced attention mechanisms. I have proposed an evolutionary framework consisting of four stages such as foundation, optimization, multi scale, and cross fusion of these channels, so as to classify the technological progress that is moving MER toward real time application in human computer interaction (HCI).1 It undertakes a systematic analysis of various fusion strategies including early, late, hybrid, and intermediate layer fusion and underscores the critical importance of techniques such as cross modal, recursive, and multi-scale attention in integrating the challenges occurring from data heterogeneity. The document outlines that Multimodal Large Language Models (MLLMs) and generative reasoning currently constitute the state-ofthe- art system providing potent instruments for explaining the hidden relations within intricate social interactions.5 Also, the survey strongly recommends integration of Explainable AI (XAI) to ensure requisite transparency and capture user confidence in emotion aware systems.7 By preparing the performance metrics which are derived from the datasets including IEMOCAP, MELD, and CMU-MOSEI, the report establishes the present performance benchmark and outlines the future research designs for the development of multimodal systems which are characterized by enhanced robustness, adaptability, and inclusivity

Fields of Study & Subjects

Emotion and Mood Recognition Sentiment Analysis and Opinion Mining Multimodal Machine Learning Applications

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN