Sentiment analysis of customer satisfaction levels on smartphone products using Ensemble Learning

Abstract

Increasingly sophisticated technological developments create new ways for people to conduct trading business. An example of this technology application is the use of e-commerce. However, there are conditions where the seller cannot measure the level of satisfaction and identify problems experienced by his customers if it is only based on the rating as the case in smartphones transactions. Therefore, a solution is needed to create a system that can filter negative and positive comments. This study offers a solution to address this issue by using machine learning employing the K-Nearest Neighbors, SVM, and Naive Bayes algorithms with hyperparameters from previous studies. This study applied the ensemble learning method with the Voting Classifier technique, which is an algorithm to combine several algorithms that have been made. From the test results, the highest accuracy was obtained by SVM with an accuracy value of 91.18% while the ensemble learning method obtained an accuracy value of 89.22%. The difference in the accuracy of training and testing for SVM and ensemble learning method is 7.1% and 4% respectively. These results indicate that the ensemble learning method can help improve the performance of sentiment analysis algorithms for comments on smartphone products.