Correlation Based Feature Selection for Movie Review Sentiment Classification

Abstract: Sentiment Analysis is one of the recent research areas in Data Mining concepts and Natural Language Processing techniques. It retrieves users or customer reviews from the web and classify the reviews using sentiment analysis approach. This paper proposes a method for sentiment classification using correlation based feature selection. First, different levels of data pre-processing techniques applied on the labeled polarity movie review dataset results in structured documents with Bag of Words. Second, correlation attribute method is used for feature selection to identify most important features. Finally, the two popular classifiers namely Naive Bayes(NB) and Support Vector Machine(SVM) are implemented and evaluated various performance measures of sentiment analysis. The proposed model concludes with the better results of accuracy using SVM classifier.

Keywords: Sentiment Analysis, Opinion Mining, Correlation, Naive Bayes, Support Vector Machine.