Abstract
Due to the enormous amount of data in online reviews, various parties, including individuals, businesses, and governments, are becoming more interested in evaluating the sentiments behind these contents. In this paper, we conducted a comparative assessment of the performance of three popular ensemble methods (Bagging, AdaBoost, and Stacking) based on five base learners (Naive Bayes, Linear Regression, Decision Tree, K-Nearest Neighbor, and Support Vector Machine) to predict sentiment classification. Experiments were performed on three different domains of online reviews including restaurant cell-phone, and movies. Results revealed that the ensembles, in general, had better performance than the individual classifiers with an average of 0.83 for precision and 0.82 for recall. When comparing the performance of three ensembles methods, Stacking (heterogeneous ensemble method) was found to be the best method, whereas Bagging (homogeneous ensemble method) recorded the lowest performance. The results offered compelling evidence that ensemble methods, especially heterogeneous with a robust classifier as a based or meta classifier, can positively improve the performance of sentiment classification.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have