Sentiment analysis of movie reviews based on NB approaches using TF–IDF and count vectorizer

Movies have been important in our lives for many years. Movies provide entertainment, inspire, educate, and offer an escape from reality. Movie reviews help us choose better movies, but reading them all can be time-consuming and overwhelming. To make it easier, sentiment analysis can classify movie reviews into positive and negative categories. Opinion mining (OP), called sentiment analysis (SA), uses natural language processing to identify and extract opinions expressed through text. Naive Bayes, a supervised learning algorithm, offers simplicity, efficiency, and strong performance in classification tasks due to its feature independence assumption. This study evaluates the performance of four Naïve Bayes variations using two vectorization techniques, Count Vectorizer and Term Frequency–Inverse Document Frequency (TF–IDF), on two movie review datasets: IMDb Movie Reviews Dataset and Rotten Tomatoes Movie Reviews. Bernoulli Naive Bayes achieved the highest accuracy using Count Vectorizer on the IMDB and Rotten Tomatoes datasets. Multinomial Naive Bayes, on the other hand, achieved better accuracy on the IMDB dataset with TF–IDF. During preprocessing, we implemented different techniques to enhance the quality of our datasets. These included data cleaning, spelling correction, fixing chat words, lemmatization, and removing stop words. Additionally, we fine-tuned our models through hyperparameter tuning to achieve optimal results. Using TF–IDF, we observed a slight performance improvement compared to using the count vectorizer. The experiment highlights the significant role of sentiment analysis in understanding the attitudes and emotions expressed in movie reviews. By predicting the sentiments of each review and calculating the average sentiment of all reviews, it becomes possible to make an accurate prediction about a movie’s overall performance.