Comparative Study of Model Efficiency for Sentiment Analysis on BRImo Reviews

This study aims to overcome the challenges in selecting the most efficient machine learning model to analyze user sentiment towards the BRImo application on the Google Play Store. User review data on BRImo, an application owned by PT Bank Rakyat Indonesia, was collected through web scraping techniques from a total of $\mathbf{1. 5 5}$ million reviews, and then 150,000 relevant samples were used. The results showed that Support Vector Machine (SVM) had the highest accuracy rate (87.13%) but required the longest training time (1310.70 seconds). In contrast, Naive Bayes had the shortest training time (0.59 seconds) but the lowest accuracy (75.00%). Logistic Regression emerged as a model that provides a balance between high accuracy and fast execution time. The contribution of this study is to provide a comprehensive guide to the efficiency and reliability of models in sentiment analysis, which can be used to assist in selecting the right model in new user review analysis.