Wilcoxon signed rank based feature selection for sentiment classification

Abstract

Sentiment analysis process is about gaining insights into the consumer’s perception using the inputs like comments and opinions shared over the web platform. Most of the existing sentiment analysis models envisaged the complexities, which is due to high volume of features notified through standard selection/extraction process. In this manuscript, the proposed solution is about using statistical assessment strategies for selecting optimal features under sentiment lexicon context. The proposed solution relies on Wilcoxon signed score for finding significance of feature towards positive and negative sentiments. Concerning to performance analysis of the proposed solution, the experimental study is conducted using benchmark classifiers like SVM, NB and AdaBoost. Results from the experimental study depict that the proposed solution can support in attaining effective classification accuracy levels of 92%, upon using less than 40% of the features too.