Comparison of Naive Bayes Algorithm and Support Vector Machine using PSO Feature Selection for Sentiment Analysis on E-Wallet Review

Abstract

Abstract Cashless payment habits have been widely applied to the transportation system, restaurants and shops in the mall area. So, it is normal if the growth of mobile payment services is currently very rapid. The ease of doing transactions and promotional offers in the form of points and cashback in digital wallet applications (e-wallets) is very beneficial for its users. One of the most popular e-wallets is OVO. With so many reviews about OVO customer opinions on social media, there has also been a lot of public opinion. These opinions can produce negative or positive statements. Sentiment analysis is the mining of opinions or text to classify opinions or user reviews, of a brand reviews, product reviews, or service reviews into the category of positive or negative opinion. The methods used in this research are Naive Bayes and SVM. Both of these algorithms are the best algorithms widely used in text classification research. However, both of these algorithms have weaknesses in several parameters. So, in this study Feature Selection is used to improve its performance. The evaluation was carried out using 10-fold cross validation. Measurement accuracy is measured by confusion matrix and ROC curves. This study uses 500 positive reviews and 500 negative reviews as data training. The results of this study indicate that the use of PSO-based Naive Bayes algorithm produces an accuracy value of 93.10 percent with an AUC value of 0.750. While the results of research from the PSO-based SVM algorithm are 91.30 percent with an AUC value of 0.970. Based on these results the accuracy value generated by the Naive Bayes algorithm is classified as Fair Classification and SVM is classified as Excellent Classification. The AUC value generated by the Naive Bayes algorithm is also smaller than SVM. Therefore, in this study found that SVM is the best algorithm in classifying text.