Performance Evaluation of Selected Deep Learning Models on Sentiment Classification of Online Movie Reviews

Abstract

The analysis of a movie's sentiment in the movie review can help understand audience's opinion and predict it's success. In this study the performance of prevalent Deep Learning (DL) models (Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM)) was compared to analyse the sentiment in reviews of online movies.
This helps in identifying an effective technique for analysis of movie reviews' sentiment and in selecting the best-suited model for practical applications. This research utilized a structured dataset which was sourced from Kaggle as the experiment data for this research. It provides a structured foundation for evaluating the effectiveness of these models in identifying sentiments accurately.
The adopted DL models were simulated in Python programming environment where recall, F1- score, accuracy, and precision are the considered metrics. The result of this was that CNN did perform much better than RNN and LSTM. The best performance was achieved on the CNN model of 84.85% accuracy, 82.47% precision, 88.81% recall and an F1-Score of 85.52% showing
more capability for sentiment classification. The RNN scored 84.18% accuracy, 84.40% precision, 84.16% recall, and 84.28% F1-Score, while the LSTM achieved 84.51% accuracy, 84.83% precision, 84.34% recall, and 84.59% F1-Score. This result advocates that CNN is a more reliable DL model for analysis of movie reviews' sentiment, offering enhanced recall and dependability in real-world
applications.