Aspect-Based Sentiment Analysis on PLN Mobile Application Using Word Embedding Bert and GRU Classification Algorithm: a Case Study of Play Store Comments in 2023

PT PLN (Persero) is an Indonesian utility provider offering electricity services via the PLN Mobile app, accessible on Android and iOS platforms. The app includes features like bill payments, token purchases, and fault reporting. Despite these services, numerous users express dissatisfaction in their Google Play Store reviews, often citing technical issues and subpar service. This study introduces a sentiment classification system utilizing the Gated Recurrent Unit (GRU) method, paired with Bidirectional Encoder Representations from Transformers (BERT) for word embedding representation. The analysis was conducted with the pre-trained indobert-base-p2 model from IndoNLU, employing a learning rate of 0.0001, 512 GRU units, a dropout rate of 0.5, 15 epochs, a batch size of 16, and the Adam optimizer. Utilizing a dataset of 7,934 reviews, the data was categorized into two sentiments: 4,422 negative and 3,572 positive reviews. The trained model achieved an accuracy of 93.39%, indicating a predominant negative sentiment in the PLN Mobile app reviews. This is because the test data has more negative than positive sentiment reviews. The performance of the sentiment analysis model across different aspects of the PLN Mobile app directly impacts key indicators by providing actionable insights into user feedback.