Table 1 A comparative study of the reviewed techniques.

Kumar et al.¹¹

A hybrid-fusion-based, innovative, and interpretable multimodal emotion detection method, VISTANet, is presented to categorize an input comprising an image, equivalent text, and speech into discrete emotion segments.

KAAP

IIT-R MMEmoRec dataset

Accuracy of 80.11%

Di Luzio et al.¹²

A novel explainable AI model to recognize vital facial movements and distinctive emotional feelings.

DNN

Extended Cohn-Kanade dataset (CK+)

–

Fu et al.¹³

A novel structure for handling inadequate conversational information in the task of MERC that deliberates the higher-order data of modality and multi-frequencies, and fully employs the semantical dependencies.

GRU and Spectral Domain Reconstruction Graph Neural Network

IEMOCAP, CMU-MOSI, and CMU-MOSEI datasets

The Time of SDR-GNN_mini is 7.52s

Li et al.¹⁴

An innovative ERNetCL model incorporates RNN- and MHA techniques in a simplified method to acquire spatial and temporal contextual data.

TE, SE, and CL

MELD, IEMOCAP, EmoryNLP, and DailyDialog dataset

Weighted-F1 of 66.31%, 69.73%, 39.71%, and 53.09% of Four Datasets

Kusal et al.¹⁵

Presents a hybrid DL model that depends on the convolutional-recurrent network employed to identify the emotions of individuals, depending on the conversational text.

Neural Network Language Model (NNLM), CNN, Recurrent Neural Network (RNN)

Empathetic Dialogues dataset

Accuracy of 73.62%

Feng et al.¹⁶

Establish a multimodal technique that combines text and speech information to capture the full benefits of emotion-relevant data, surpassing the application of a multiscale MFCC multiview AM.

Multiscale MFCC and Multi-Head Attention (MHA)

IEMOCAP and MSP-IMPROV dataset

IEMOCAP is 0.754 in WA and 0.742 in UA

Zhang et al.¹⁷

An automated emotion study method is utilized to allow the machine to understand the emotional inference transferred by the person’s EEG signals.

CNN and LSTM

DEAP dataset

Accuracy of 92.98%

Omarov and Zhumanov¹⁸

Projects an innovative Bi-LSTM technique for emotion study and identification in textual content, able to acquire either preceding or upcoming context to enhance performance.

Bi-LSTM

Kaggle Emotion Detection Dataset

Weighted Average of 90% Precision, 90% Recall, and 90% F-Score

Hicham and Nassera¹⁹

To develop a stacked DL technique for efficient multilingual opinion mining.

RoBERTa-GRU, RoBERTa-LSTM, RoBERTa-BiGRU, RoBERTa-BiLSTM, Adam optimizer, AEDA, SMOTE, GPT

French, English, Arabic, Cohen’s kappa, ROC-AUC, Accuracy, MCC, K-fold

High Efficacy, Improved Classification

Mahajan, More, and Shah²⁰

To develop and evaluate models for recognizing single and mixed emotions in multilingual, code-mixed YouTube comments.

LR, SVM, NB, RF, LSTM, BiLSTM, GRU, CNN

13,000 multilabel YouTube comments, Accuracy, F1 score

SVM Highest Accuracy

Zhu et al.²¹

To develop an accurate and efficient medical question-answering system using advanced AI techniques.

Knowledge Embedding, Transformer-based Architecture, Knowledge Understanding Layer, Answer Generation Layer

MCMLE and USMLE Datasets

82.92% on MCMLE and 64.02% on USMLE

Khan et al.²²

To develop an efficient DL method for accurate violence detection in surveillance videos using a two-stream approach.

Two-Stream Architecture, 3D Convolution Network, Background Suppression, Optical Flow Analysis, Depth-Wise 3D Convolutions

RWF2000, RLVS

Accuracy, Efficiency

Arumugam et al.²³

To develop a novel multimodal by integrating audio, visual, and text inputs for accurate emotion recognition.

AVTEFN, Hybrid Wav2Vec 2.0 + CNN, BERT with Bi-GRU, Attention-Based Fusion

Benchmark Dataset

Accuracy at 98.7%, Precision at 98.2%, Recall, at 97.2%, and F1-score of 97.49%

Khan et al.²⁴

To enhance multimodal emotion recognition by capturing inter- and intra-modal relationships using a joint transformer-based model.

MER, JMMT

IEMOCAP, MELD

Accuracy, F1-Score

Alyoubi and Alyoubi²⁵

To develop an optimized transformer-based multimodal emotion recognition framework for accurate emotion classification.

BERT/RoBERTa for Text, wav2vec 2.0 for Speech, ResNet50/VGG16 for Visuals, Cross-Modal Attention, SHAP Explainability

Multimodal EmotionLines (MELD) Dataset

Accuracy, Explainability

Vani et al.²⁶

To develop and evaluate Text Fusion + through advanced text analysis and audio output.

OCR, NLP, TTS, DL Summarization, NLP-based Q&A Module

Standard Dataset

Summarization Accuracy, User Accessibility

Khan et al.²⁷

To explore sequence learning techniques for accurate auditory emotion recognition using advanced RNNs.

LSTM, GRU, Bi-Directional LSTM, Deep/Multilayer Architectures

Emotion Dataset

Accuracy, Model Robustness

Ghous, Najam, and Jalal²⁸

To detect emotional states in individuals with cognitive disabilities using EEG data and advanced ML models.

BF, Downsampling, AAFST, Multi-Class SVM

SEED-IV

Accuracy, Emotion Detection

Patil et al.²⁹

To predict learning disabilities using handwritten text analysis and intelligent systems.

DNN, Character Confusion Detection, Pattern Extraction Techniques

IAM Handwriting Dataset

Accuracy, Scalability

Mishra et al.³⁰

To develop a transformer-based NLP model to support mental health monitoring.

Transformer Architecture, GPT-4 and BERT, Adam Optimizer

English Twitter Dataset

94% Accuracy