Sentiment Analysis on WeTV Application Reviews Using Naïve Bayes: A Study of Preprocessing, Balancing, and Model Performance

Abstract

This study investigates the application of the Naïve Bayes classification algorithm for sentiment analysis of user-generated reviews on the WeTV application available on the Google Play Store. A structured methodology was employed, consisting of data scraping, sentiment labeling based on heuristics, multi-stage preprocessing, class balancing using Synthetic Minority Over-sampling Technique (SMOTE), and performance evaluation through standard metrics. Prior to balancing, the model exhibited strong performance on the dominant class but underperformed on the minority class. The introduction of SMOTE led to improved F1-scores, particularly for positive sentiment, increasing from 61% to 64%, while maintaining overall accuracy around 71%. These findings confirm that Naïve Bayes, when supported by effective preprocessing and data balancing, can deliver robust and interpretable classification results in text mining tasks. This research contributes to the growing literature on machine learning for opinion mining and provides practical implications for developers aiming to extract structured insights from large-scale user reviews.