A Performance Evaluation of Sentiment Classification Applying SVM, KNN, and Naive Bayes

Abstract

The rising use of the internet and social networks has opened up new avenues for individuals to express themselves. It’s also a platform with a plethora of information where an individual can see other people’s thoughts, which are diverged into numerous sentiment categories and are slowly becoming a primary part of the decision. This study makes a significant contribution to sentiment classification, which is effective in determining data in a big amount of tweets with de-contextualized sentiments which are often positive or negative, or in the middle. To accomplish this, we initially pre-processed the raw data, and then draw out the meaningful words and phrases (characteristic vector), then picked the characteristic vector list, and then applied machine-learning classification methods including Naive Bayes, KNN, and SVM. And at last, we assessed the classifier’s performance using the terms recall, accuracy, and precision, as well as the F1-score. Support Vector Machine has the highest accuracy of 92 percent, followed by KNN and Naive Bayes with 88 and 85 percent accuracy, respectively.