Sentiment Analysis of Tokopedia Application Reviews on Google Play Store Using BERT

Tokopedia, one of Indonesia’s largest e-commerce platforms, receives millions of user reviews on Google Play Store daily. This research aims to analyze the sentiment of user reviews on the Tokopedia application by utilizing the Bidirectional Encoder Representations from Transformers (BERT) model. A total of 5,200 reviews were collected through scraping techniques using the 'google-play-scraper' module. Once collected, the review data was processed through the stages of cleaning, normalization, stopword removal, tokenization, and stemming to improve the quality of analysis. The sentiments of the reviews were then categorized into positive, negative, or neutral. The results show that 38.5% of the reviews are negative, 38.4% are positive, and 23.1% are neutral, reflecting a reasonably balanced sentiment distribution. The data was divided into training, validation, and testing sets to optimize the model performance. The BERT model achieved high performance, with 89% accuracy, 88.90% precision, 88.89% recall, and 88.83% F1 score. These results provide valuable insights for Tokopedia to improve service quality and enhance app user experience.