Product Backlog Recommendations Using Sentiment Analysis with Topic Modelling and Text Summarization in the XYZ E-Wallet Application

This research investigates sentiment analysis of user reviews for the XYZ e-wallet application on the Google Play Store, which currently holds a low rating of 2.7 out of 5. A preliminary study, including a focus group discussion with key stakeholders, revealed that this rating is primarily due to insufficient consideration of user feedback during the Software Development Life Cycle (SDLC). This study employs word2vec and Long Short-Term Memory (LSTM) models to analyze user sentiments, using a dataset comprising reviews collected between July 1, 2023, and June 30, 2024. The developed sentiment analysis model achieved an accuracy of 91.40%. Subsequently, the model was applied to predict sentiments from new review data spanning July 1, 2024, to July 31, 2024. The results were visualized using BERTopic for topic modeling and BART for text summarization. Insights from these visualizations informed recommendations for application improvement, particularly addressing frequent issues with balance top-ups that fail to update, which were incorporated into the product backlog to enhance user satisfaction. This study is limited to a single sentiment analysis model, topic modeling technique, and summarization method. Future research should explore applying the proposed model to other applications and evaluating alternative algorithms for improved performance.