Sentiment Analysis of Money Lover App Reviews using Random Forest and Naïve Bayes

Abstract

This study aims to analyze user sentiment toward the Money Lover application and to compare the performance of two different machine learning algorithms, Random Forest and Naïve Bayes, in binary classification of review data. A total of 3,000 comments were collected using web scraping techniques and then classified into positive and negative sentiment categories. The preprocessing stage included text cleaning, normalization, tokenization, stopword removal, and stemming. In the next stage, term weighting was performed using TF-IDF to convert the text into numerical vector representations. The results provide insights into the overall sentiment tendencies of users toward the Money Lover application and demonstrate the effectiveness of both algorithms in processing textual reviews within the financial domain. Based on model evaluation, the Random Forest algorithm achieved superior average performance, with an accuracy of 94%. Meanwhile, the Naïve Bayes algorithm showed slightly lower performance, achieving an accuracy of 92%. These findings were supported by cross-validation results and ROC curve analysis, which indicated that Random Forest consistently outperformed Naïve Bayes. The performance difference suggests that an ensemble-based approach such as Random Forest is better able to handle textual variation in review data, resulting in more stable and accurate sentiment classification.