Merchant Review Analysis in Food Delivery Apps: LSTM-Based Sentiment and Dual-Model Topic Classification with LLM and Logistic Regression

Technological advancements, particularly in natural language processing and deep learning, have enabled the automatic extraction of insights from user-generated content such as user reviews. In the context of food delivery services, merchant applications like GoBiz, Shopee Partner, and Grab Merchant offer various features and policies that often cause confusion and dissatisfaction among business users. This study proposes a sentiment analysis and topic classification approach using user reviews from the Google Play Store to help merchants evaluate these merchant applications. The system employs a Long Short-Term Memory (LSTM) model for sentiment analysis and uses logistic regression and zero-shot for topic classification. Data preprocessing involved data reduction, data labeling, and text preprocessing, followed by data augmentation through rule-based and generative paraphrasing techniques. The sentiment model for GoBiz achieved the highest accuracy of 92.55% and a weighted F1-score of 94%, while the Shopee Partner model produced the most contextually relevant sentiment predictions. The system also categorized topics into key areas such as service, technical issues, platform policies, etc., providing meaningful insights to support application evaluation and future improvement.