An application of MOGW optimization for feature selection in text classification

Due to extensive web applications, sentiment classification (SC) has become a relevant issue of interest among text mining experts. The extensive online reviews prevent the application of effective models to be used in companies and in the decision making of individuals. Pre-processing greatly contributes in sentiment classification. The traditional bag-of-words approaches do not record multiple relationships among words. In this study, emphasis is on the pre-processing stage and data reduction techniques, which would make a big difference in sentiment classification efficiency. To classify opinions, a multi-objective-grey wolf-optimization algorithm is proposed where the two objectives aim for decreasing the error of Naive Bayes and K-nearest neighbour classifiers and a neural network as the final classifier. In evaluating this proposed framework, three datasets are applied. By obtaining 95.76% precision, 95.75% accuracy, 95.99% recall, and 95.82% f-measure, it is evident that this framework outperforms its counterparts.