Using SVM and KNN for Predicting Customer Response Sentiment of M-PAJAK Application

Abstract

M-Pajak, an application initiated by the Directorate General of Taxes, signifies the modernization of taxation and serves a crucial function. This application facilitates taxpayers in meeting their tax obligations. User satisfaction with this application may be assessed via reviews on the Google Play Store. While this application fulfills client satisfaction, its sustained success is significantly contingent upon user contentment and experience. Sentiment analysis is essential for elucidating user evaluations and interactions with the program. This research analyses the sentiment of M-Pajak application reviews on Google Play using Support Vector Machine (SVM) and K-Nearest Neighbour (KNN), supported by the Term Frequency-inverse Document Frequency (TF-IDF) feature extraction method. A total of 1000 reviews between December 11, 2022 and December 2, 2023 were processed using KNN and SVM. The KNN algorithm yielded 153 positive predictions and 847 negative predictions and achieved 94% of accuracy. Meanwhile, SVM achieved an accuracy of 88.10%, with 325 positive predictions and 675 negative predictions. The results demonstrate the superiority of KNN in sentiment classification of M-Pajak reviews. This study also indicates that negative comments outnumber positive ones in this application. This serves as a signal for the Directorate General of Taxation to enhance user satisfaction with the M-Pajak application through continuous updates.