Comparative Analysis of Sentiment Classification on IMDB 50k Movie Reviews: A Study Using CNN, LSTM, CNN-LSTM, and BERT Models

Sentiment analysis of movie reviews plays a critical role in understanding audience perception and informing marketing strategies in the film industry. This study investigates the effectiveness of various deep learning models for sentiment classification on the IMDB 50k movie review dataset and compares the performance of Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, a combined CNN-LSTM architecture, and pre-trained transformer models like BERT and RoBERTa, coupled with Deep Neural Networks (DNNs). The research explores the impact of pre-processing techniques like text cleaning, tokenization, and embedding on model performance. Hyperparameter optimization is employed to fine-tune each model for optimal accuracy. Metrics such as precision, accuracy, recall, and Fl-score are employed to evaluate the performance of each model in categorizing movie reviews into positive or negative sentiments. The findings reveal that RoBERTa+DNN achieved the highest performance with an accuracy of 0.9205, followed closely by BERT+DNN at 0.9135. While CNN and LSTM models achieved good accuracy (0.8847 and 0.8602 respectively), they were surpassed by the combined CNN-LSTM architecture (0.8851) and the pre-trained transformer models. It identifies RoBERTa+DNN as the most effective model for classifying IMDB movie reviews, achieving the highest Fl-score (0.9217). This study contributes through the integration of BERT and RoBERTa models with Dense Nets to enhance sentiment classification results, alongside comprehensive comparisons with various es-tablished classification models. The research also emphasizes the importance of pre-processing techniques and hyperparameter optimization in achieving optimal model performance.