Analysis Comparison of Hard and Soft Voting Ensemble Model for Sentiment Analysis on IMDB Movie Reviews

In the context of movie reviews, sentiment analysis plays a crucial role as these reviews often capture the emotions and opinions of viewers regarding the film. With the growing number of reviews across platforms like social media and movie review sites, it becomes feasible to automatically classify them into positive or negative sentiments. However, manually analyzing a dataset of 50,000 reviews is impractical. Therefore, this study focuses on conducting sentiment analysis on the IMDB movie review dataset, consisting of 50,000 entries, using an ensemble voting model. The machine learning models applied in this ensemble include Random Forest (RF) and Adaptive Boosting (AdaBoost). Two test scenarios were employed: the first compared the soft and hard voting approaches, while the second examined the impact of different training-test data splits. The findings indicate that the soft voting ensemble (SVE) model with an 80%-20% training-testing split achieved an accuracy of 88%, an F1-score of 88%, and an AUC of 0.94, which is 4% higher than the hard voting model.