Evaluasi Kinerja Algoritma Naïve Bayes, SVM, dan IndoBERT pada Analisis Sentimen Ulasan Pengguna Gojek Berbasis Text Mining

Abstract

This study aims to evaluate and compare the performance of Naïve Bayes, Support Vector Machine (SVM), and IndoBERT algorithms in the task of sentiment classification of user reviews on the Gojek application. Data were collected through web scraping from the Google Play Store and subsequently labeled into three sentiment categories: negative, neutral, and positive. A quantitative approach with a descriptive-comparative design was employed. The research procedure consisted of data collection, text preprocessing, dataset splitting into training and testing sets, model development, and evaluation using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results indicate that the IndoBERT algorithm achieved the best performance, with an accuracy of 92.46%, outperforming Naïve Bayes (87.94%) and SVM (86.43%). Furthermore, IndoBERT demonstrated greater consistency in precision, recall, and F1-score across all sentiment categories. In contrast, Naïve Bayes exhibited a tendency to misclassify certain classes, while SVM showed relatively stable performance, although it did not reach optimal results. These findings suggest that transformer-based approaches are more effective in capturing the contextual complexity of the Indonesian language. This study contributes by providing a comparative analysis of classical and transformer-based methods in Indonesian sentiment classification and offers empirical evidence of the superiority of transformer-based approaches in capturing linguistic contextual nuances in user reviews of digital applications.