Sentiment Analysis of Wargaku Surabaya Apps Reviews using Naive Bayes and Support Vector Machine (SVM) Methods

Abstract

Digital public service applications in Indonesia are increasingly used to improve citizens access to government services, generating large volumes of feedback that are difficult to analyze manually. Moreover, many previous studies focus on polarity-based sentiment, which may not adequately capture specific user emotions. This study analyzes feedback on the Wargaku Surabaya application by classifying emotions into five categories: anger, disappointment, sadness, pride, and happiness. A total of 1,406 texts were collected (2021–2025), with 1,386 retained after preprocessing. Data were primarily sourced from Google Play Store reviews, supplemented by comments from Threads and YouTube. The research employs text preprocessing, TF-IDF weighting, and lexicon-based labelling with the generated labels reviewed on a subset of the dataset before model training. Emotion classification was performed using Naive Bayes (NB) and Support Vector Machine (SVM), evaluated via a train–test split and confusion matrix. Results show that SVM achieved 84% accuracy, 85% precision, 84% recall, and an 84% F1-score, outperforming NB with 58% accuracy. These findings indicate that SVM is more reliable for multi-class emotion classification in digital public services.