Single Aspect and Multiple Aspect Sentiment Analysis for Movie Reviews

Abstract

The use of machine learning is increasing rapidly in various domains including the film industry. Reviews from the viewers is very important for the production houses. Since the world is changing rapidly, the opinion of the viewers are changing rapidly. It is essential to keep track of different type of viewers and fans, in order to understand their opinions to improve the content for the next film. This work proposes a novel sentimental analysis technique to learn the opinions of the viewers from IMDb. Two different feature extraction techniques: Bag of Words (BoW) and Term Frequency & Inverse Document Frequency (TF-IDF) is used, and three classification algorithms: Logistical Regression, Naïve Bayes, and Sequential Gradient Descent (SGD) is used. These six combinations are implemented, and the performance are compared. The TF-IDF model worked better than the BoW model, while the SGD model had the highest accuracy among the classification techniques with an accuracy of 82.07%. Multiple aspect sentimental analysis is also performed for identifying multiple aspects within the same review, and this was performed efficiently.