Sentiment Analysis of User Reviews on Maxim Application Using the Long Short-Term Memory (LSTM) Methods

Abstract

The technological developments have encouraged the emergence of app-based transportation services that are increasingly popular with the public, one of which is the Maxim app. Despite offering convenience in booking transportation and other services, this app still receives various reviews from users regarding service quality. User feedback is provided through the Maxim app review section available on the Google Play Store platform. Sentiment analysis is applied in this study to identify shortcomings in the Maxim app, to help developers improve service quality and understand user satisfaction. The research procedure it comprises several phases, including data collection, text preprocessing, determining sentiment labels, assigning weights to terms, and a classification process using the Long Short-Term Memory (LSTM) algorithm. This studi unlike previous studies that commonly used classical machine learning techniques including Naïve Bayes and SVM, or BiLSTM, this research applies an LSTM model with lexicon-based sentiment labeling to improve consistency and contextual understanding in sentiment classification. A confusion matrix was utilized to evaluate the model’s performance. Overall, 1,200 user reviews were gathered through web scraping techniques from June 2024 to June 2025. The sentiment classification process uses a lexicon-based method to categorize user reviews grouped into three sentiment classes: positive, neutral, and negative. The findings suggest that 762 reviews are labelled as positive, 157 as neutral, and 281 as negative. The LSTM method testing demonstrated excellent performance, achieved 95.21% accuracy, 97.22% precision, 84.02% recall, and an F1-score of 88.84%.