Comparative Study of Machine Learning Models for Classifying Sentiment in Google Gemini App Reviews

This study uses user review data collected from the Google Play Store with the data collection limited up to September 11, 2025, and restricted to reviews written in the Indonesian language. The data was obtained through a web scraping method using Python executed on Google Colab, enabling systematic extraction of reviews along with supporting attributes such as rating scores and review dates. The objective of this research is to compare the performance of the Naive Bayes, Support Vector Machine (SVM), and Random Forest models in sentiment classification tasks on the Google Gemini application based on evaluation metrics. To ensure data quality, a validation process was conducted by removing duplicate entries and filtering irrelevant or incomplete reviews. The dataset then underwent preprocessing, including case folding, tokenization, stopword removal, and stemming to standardize the textual data and reduce noise prior to modeling. After all cleaning and preprocessing stages were completed, a total of 12,823 clean reviews were obtained and used as the final dataset in this study. The processed data enables meaningful interpretation of user sentiment patterns toward the Google Gemini application and can be utilized to evaluate classification model performance as well as to provide insights into user perceptions as a basis for drawing research conclusions.