Sentiment Analysis on Bengali Movie Reviews using Multinomial Naïve Bayes

Opinion mining of consumers has become one of the undeniable approaches for finding gaps in businesses’ marketing strategies. For today’s gently growing film industry of Bangladesh, sentiment mining of viewers feedback on their specified work has become inevitably a dire need for the production company and also for the audiences to take decision about watching a film. With the availability of such a great extent of e-content on film reviews, it becomes handy to analyze the viewers’ sentiment on any film. Because of the lack of structured research work on Bengali movie review sentiment analysis, we are interested to focus this kind of research work. We collect 3000 Bengali movie reviews and extract TF-IDF features by using uni-gram, bi-gram, and tri-gram models. We use several machine learning classifiers for performing classification solutions on extracted TF-IDF features of the corpus. Experimental results show that the Multinomial Naïve Bayes classifier provides the highest accuracy, 86%, on uni-gram features of the validation data.