Sentiment Analysis of PLN Mobile Review Data Using Lexicon Vader and Naive Bayes Classification

The PLN Mobile application was introduced by PLN as a superior digital platform to suit all client needs, provide convenience, and give a unique power service experience. The PLN Mobile application offers simplicity and speed of service to PLN customers with around 24.8 million users and more than 35 million registered customer IDs (https://web.pln.co.id/media/press-release/2022/07, July 2, 2020). The customer ratings for the Google Play Store app range from 1 to 5. The quality of the program is not accurately portrayed when users provide ratings that do not correspond to their reviews. It will take some time to read all of the reviews on the PLN Mobile app because there are so many of them. Classification is used to measure public sentiment. The sentiment research used 1,000 review sample data points from 67950 population data points obtained from the PLN Mobile application between January and June 2022. Initially, online scraping, machine translation, data labelling, text pre-processing, TF-IDF, text classification, and model evaluation approaches were used to acquire review data. The labelling results for the Lexicon-based text classification approach with the Vader Lexicon dictionary-based approach are 489 positive sentiments, 145 negative sentiments, and 366 neutral. Positive classes get a rating of 67%, neutral classes get a rating of 6%, and negative classes get a rating of 27%. Based on the comparison of positive, neutral, and negative classes from the data sample, PLN Mobile application users give ratings that do not match the reviews given. The Naive Bayes approach is also utilized in the categorization process. For the distribution of test and training data, the author employs a split data ratio of 90:10. The confusion matrix evaluation technique yields 70% accuracy with the F1 score value for each class, namely for the positive class (62%), neutral class (57%), and negative class (77%).