Novel Framework for Sentiment Analysis of Movie Reviews based on Large Language Models and Deep Learning Methods

Background: Sentiment analysis has become a crucial research area, enabling businesses to gauge public opinion on products or movies without manually reviewing content. Methods: Using IMDb data that was heavily preprocessed to improve model training, this research investigates sentiment analysis of movie reviews using the ELECTRA model. A multimodal approach incorporating textual, audio, and visual data was explored to improve emotional context understanding. Results: ELECTRA outperformed other models, including BERT, SVM, and CNN, achieving an impressive 95.6% accuracy based on F1score, precision, recall, and accuracy metrics. While CNN also performed well with 90% accuracy, BERT and SVM achieved 84.01% and 86.18%, respectively, demonstrating comparatively lower effectiveness. Conclusion: The findings highlight that deep learning models, particularly ELECTRA, excel in extracting subjective sentiments from movie reviews. As large language models continue to evolve, sentiment analysis methodologies will likely advance further, offering even greater accuracy and contextual understanding in the future.