Sentiment Analysis of Public Satisfaction with the 'INFO BMKG' Application using Naive Bayes, SVM, and KNN

Abstract

This study aims to analyze public sentiment regarding the Info BMKG application on the Google Play Store. With the increasing number of users of information-based applications, understanding how users perceive and evaluate such applications has become essential. This research employs three classification algorithms—Naive Bayes, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN)—to categorize user reviews into positive, neutral, or negative sentiments. The dataset was obtained through web scraping from the Google Play Store, consisting of usernames, dates, star ratings, and user comments. Data preprocessing was conducted to clean and prepare the data for analysis. Additionally, a web-based data mining application was developed to facilitate data processing and result visualization. The study aims to present the distribution of sentiment (positive, neutral, and negative) toward the Info BMKG app and help developers understand the factors that influence user satisfaction. Moreover, it contributes to the field of sentiment analysis and information technology, particularly in disaster-related information applications. Based on model evaluation, the Naive Bayes algorithm demonstrated the best performance with an accuracy of 79.84%, precision of 60%, recall of 58%, and the fastest runtime at 0.19 seconds. KNN achieved an accuracy of 74.35% with the lowest recall at 44%, while SVM had an accuracy of 72.26% but required the longest runtime at 611 seconds. AUC validation further confirmed the superiority of Naive Bayes, with the highest scores across all sentiment categories. Thus, Naive Bayes is shown to be the most optimal for sentiment analysis in this study, whereas KNN and SVM showed certain limitations, particularly in efficiency and classification accuracy.