Sentiment Analysis in Movie Reviews Using Document Frequency Difference, Gain Ratio and Kullback-Leibler Divergence as Feature Selection Methods and Multi-layer Perceptron Classifier

Abstract

Both industries and people are interested in knowing the reviews of the products or movies. If the number of reviews are large, manual classification of reviews into either positive or negative classes, is tedious and time consuming. As an alternative, Sentiment Analysis, a subdomain in Natural Language Processing aims to automate the above process, by training the models by using reviews as the training data. These are popular because this is very essential for making decisions by knowing the other person’s opinion. In this paper, Document Frequency Difference, Gain Ratio and Kullback-Leibler divergence are used for feature selection and the classification is done with Multi-Layer Perceptron classifier. This is done with movie reviews dataset like SAR14, IMDB11, Cornell open source benchmark dataset. The results show that Document Frequency Difference and Gain Ratio and feature selection methods have better sentiment classification performance i.e. with better accuracy and reduced error.