<![CDATA[The growth of electronic commerce (e-commerce) in Indonesia has led to an increase in user review volumes on digital platforms, including the Tokopedia application on the Google Play Store. The large number of reviews makes manual analysis inefficient and difficult to perform quickly for service evaluation purposes. This study develops a sentiment classification system by integrating Naïve Bayes, TF-IDF, and PySastrawi in a Flask-based web dashboard for real-time prediction. The method used is the Naïve Bayes algorithm with Term Frequency–Inverse Document Frequency (TF-IDF) feature extraction and five stages of text preprocessing using PySastrawi, namely case folding, cleaning, tokenizing, stopword removal, and stemming. The dataset was collected through a Google Play Store scraping process consisting of 5,000 reviews. A total of 278 three-star reviews were removed due to semantic ambiguity, resulting in a final dataset of 4,722 reviews consisting of 2,424 negative and 2,298 positive sentiments. The data were split using an 80:20 ratio into 3,777 training data and 945 testing data. The experimental results show an accuracy of 86.14%, precision of 87.18%, recall of 86.14%, and F1-score of 86.00%. For the negative class, precision and recall were 80% and 98%, respectively, while for the positive class, precision and recall were 97% and 75%, respectively. This study produces an integrated web dashboard system capable of presenting data visualization, model evaluation, and real-time sentiment prediction directly through a web interface.]]>
Copyrights © 2026
