Term Co-occurrence Based Feature Selection for Sentiment Classification

In this paper, the strategy of feature selection for sentiment classification explored and compared to other significant feature selection strategies found in contemporary literature. The feature selection models performed using the statistical measure of t-score and z-score. SVM, NB and AdaBoost classifiers used for classification and compared. The objective of the paper is to explore and evaluate the scope of statistical measures for identifying the optimal features and its significance to classify the opinion using divergent classifiers. Performance analysis carried out on varied datasets with diverse range like the movie reviews, product reviews and tweets, the experiments carried out on feature selection strategies proposed and other strategies found in literature. From the results of the experimental studies, it is evident that optimal features selected using t-score and z-score are robust and outperformed the other feature selection strategies. In order to assess the significance of the feature selection models proposed, the classification process carried out using three classifiers called SVM, NB and AdaBoost. The classification accuracy about the features obtain by proposed models is much higher that compared to the classification accuracy obtained for the features selected by other contemporary models. Among the three classifiers that used to assess classification accuracy, AdaBoost has outperformed the other two models of SVM and NB.