Sentiment Analysis of User Reviews for the PLN Mobile Application Using Naïve Bayes and Long Short-Term Memory

This study explores large-scale sentiment analysis of user reviews for the PLN Mobile application to better understand public perception and provide quantitative insights for improving digital electricity services in Indonesia. Addressing the lack of benchmarks for Indonesian public-service apps—where prior studies rely on smaller datasets and traditional machine learning—this research positions sentiment analysis as a tool for continuous user experience monitoring. A total of 50,000 Indonesian-language reviews from Google Play were collected and pre-processed using cleaning, case folding, tokenization, stopword removal, normalization, and stemming. Sentiments (positive, neutral, negative) were assigned using a domain-specific Indonesian sentiment lexicon, yielding approximately 40% positive, 35% neutral, and 25% negative labels. Two models were applied: Multinomial Naïve Bayes using TF-IDF features and a Long Short-Term Memory (LSTM) model with 100-dimensional word embeddings and a 128-unit LSTM layer. Naïve Bayes achieved 70.89% accuracy (F1-score: 0.6964), while LSTM outperformed it with 98.02% accuracy (F1-score: 0.9800). These results highlight the superiority of deep learning in sentiment monitoring and offer a scalable framework to help PLN and policymakers enhance digital public service delivery.