Fine-Tuning IndoBERTweet for Sentiment Analysis of Indonesian Restaurant App Reviews: A Case Study of McDonald's Indonesia
Abstrak
The increasing adoption of mobile applications in the restaurant industry has generated large volumes of user reviews that provide valuable insights into customer satisfaction and application performance. However, manually analyzing thousands of reviews is inefficient, highlighting the need for automated sentiment analysis techniques. This study aims to develop and evaluate a fine-tuned IndoBERTweet model for sentiment analysis of user reviews of the McDonald's Indonesia mobile application on Google Play Store while identifying dominant complaint patterns that can support application improvement. A total of 4,000 Indonesian-language reviews were collected using Google Play Scraper. After removing boycott-related reviews and performing text preprocessing, 3,897 reviews were retained for analysis. Sentiment labels were automatically assigned based on star ratings, and the training dataset was balanced using Random Oversampling and class-weighted optimization. The IndoBERTweet model was fine-tuned using an 80:20 train–test split for three training epochs in Google Colaboratory. Model performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis, while negative reviews were further examined through unigram and bigram frequency analysis. The proposed model achieved an accuracy of 85.38%, with a weighted precision of 88.16%, a weighted recall of 85.38%, and a weighted F1-score of 86.64%. Negative sentiment was classified most accurately (F1-score = 91.72%), whereas neutral sentiment remained challenging because of severe class imbalance. Complaint analysis revealed that the most common user issues involved login authentication, application errors, slow performance, device compatibility, promotions, and loyalty-point redemption. These findings demonstrate that fine-tuned IndoBERTweet is effective for analyzing Indonesian restaurant application reviews and can provide actionable insights for improving digital service quality and user experience.
Identifier
- DOI: https://doi.org/10.63956/jitar.v2i1.75
- OpenAlex: https://openalex.org/W7171855802