Comparing different classifiers and feature selection techniques for emotion classification

Abstract

In this study, we have explored the potentiality of six different supervised machine learning approaches and feature selection filters to recognize four basic emotions (anger, happy, sadness and surprise) using three different heterogeneous emotion-annotated dataset which combines sentences from news headlines, fairy tales and blogs. For classification purpose, we have chosen the feature set to include the bag-of-words. Our study reveals the fact that the use of the resampling filter and the feature selection filters together contribute towards boosting the prediction accuracies of the classifiers. However, the boosting capabilities are more profound in the resampling filter in comparison to the use of the different feature selection filters.