Author(s): Nisha

Abstract: Aiming at binary sentiment categorisation, this study compares and contrasts four deep learning models—LSTM, CNN, BiLSTM, and BERT—using the IMDb movie review dataset. Finding the optimal model for assessing and classifying the intricate natural language phrases used in user evaluations is the main objective. Some of the preprocessing procedures were lowercasing, noise removal (From HTML components or punctuation), tokenisation, lemmatisation, stop-word removal, padding, and truncation to normalise length. Half of the 50,000 tagged reviews were good, and the other half were negative. The dataset was subsequently made public on Kaggle. Exploratory data analysis (EDA) made use of class distribution plots, word clouds, histograms, and common word bar charts to shed light on the structure and sentiment-rich language of the dataset. The cleaned data was used to train and evaluate the chosen models. A number of performance metrics were employed, including F1-score, recall, accuracy, and precision. With an impressive 99.81% F1-score, 99.78% recall, 99.82% accuracy, and 99.84% precision, BERT stood out as the top model among all the choices. Its superior performance is due to its ability to identify semantic subtleties more efficiently, which is made possible by its bidirectional focus mechanism and contextual embeddings. Since sentiment analysis and other real-world applications rely on precise text interpretation, BERT is a great pick.

DOI: 10.33545/27075907.2025.v6.i1a.86

Pages: 56-63 | Views: 446 | Downloads: 179

Download Full Article: Click Here