KINERJA MULTINOMIAL NAÏVE BAYES PADA ANALISIS SENTIMEN ULASAN PENGGUNA APLIKASI MOBILE JKN

Abstrak

The Mobile JKN application is a key digital service provided by BPJS Kesehatan to support access to national health insurance. User reviews on Google Play Store contain rich information about satisfaction and technical issues, but these data are unstructured and difficult to interpret at scale. This study aims to evaluate the performance of a Multinomial Naïve Bayes model for classifying sentiment in Indonesian-language reviews of the Mobile JKN application. Approximately 10,000 recent reviews were collected via web scraping using the google-play-scraper library and processed through several text preprocessing stages, including cleaning, case folding, tokenization, stopword removal, and stemming. The sentiment labels (positive, negative, neutral) were automatically derived from the rating scores using a distant supervision approach. Text features were represented using TF–IDF and used to train and test a Multinomial Naïve Bayes classifier. Model performance was evaluated using accuracy, precision, recall, and F1-score, complemented by a confusion matrix, sentiment distribution visualization, and wordclouds for each sentiment class. The results show that the model achieves good overall accuracy and performs particularly well in identifying positive and negative sentiments, while the neutral class remains more challenging due to shorter review length and semantic overlap with the other two classes. Sentiment distribution indicates that negative reviews still dominate, highlighting persistent technical issues such as login failures, verification problems, and application errors. These findings demonstrate that Naïve Bayes combined with TF–IDF is effective for large-scale sentiment analysis of user feedback on public service applications.