Extremely Randomized Tree Based Sentiment Polarity Classification on Online Product Reviews

The sentiment analysis of user reviews on social networking is one among the fundamental process carried out by online business organizations to improve the quality of their products and retain the customers and thereby lift the monetary benefit. Although numerous analysis models exist, still there is space for improving the performance and accuracy of informal text data-based classification models. In this examination, we conduct a comparative assessment of the effectiveness of unigram feature set extricated utilizing n-gram technique with three ensemble methods namely Extremely Randomized Tree, Voting, and Bagging classifier based on the following five baseline classifiers Random Forest, Naive Bayes, K-NearestNeighbor, Ridge Classifier, and Support Vector Machine to identify polarity from mobile product reviews. Among the three ensemble methods, the Extremely Randomized Tree technique has better outcomes with the accuracy of 98% for positive and 85% for negative cases, with an overall accuracy of 96.8%. The error rate of all the three ensemble classifiers is also under 0.5% which uncovers that ensemble classifiers performs better compared to individual classifiers.