A Novel Feature Reduction Method in Sentiment Analysis

Abstract

With the genesis of the Internet and the world wideweb, we have seen an enormous growth of data and informationon the web, as well as an increase in digital or textual opinions,sentiments and attitudes that have been remarked upon inreviews. More reviews in document-level have expressed a highdimensionalin feature space. The main task of feature selectionand feature reduction is a reduction dimension in feature spacewhile, at the same time, ensuring that is no loss in the minimumof accuracy. There are several factors to consider in reductiondimension of a term - document matrix of feature space. It canlead to removal of irrelevant and useless features; including as aresult, more efficient categories, easier analysis more accuratelyof sentiment after reduction. For this aim, we have proposed anovel feature reduction method using standard deviation basedon more variation or dispersion of features in feature space. Weused three popular classifiers, namely: Naive Bayes, MaximumEntropy and Support Vector Machine for sentiment classificationand ensemble of these classifiers. We then compared ourproposed method with other feature reduction methods used onbook and music reviews. Results show that classification by usingthe novel method improved the accuracy of sentimentclassification.