Sentiment Classification of Text Reviews Using Novel Feature Selection with Reduced Over-fitting

Abstract

Sentiment Classification is an important and hot current research area. This extended abstract of our work observes the effect of some machine learning algorithms like Naive Bayes, SVM and their variants on the movie review data. We have used a novel and hybrid feature selection/reduction technique which is minimizing the number of features exponentially. The results show that with our feature selection procedure there is an improvement in classification efficiency compared to the previous work and with reduced over-jitting.