Development of Sentiment Analysis System of Simple Pol Application on Google Play Store Using Naive Bayes Classifier Method and BERT Prediction

Abstract

Digitalization in public services raises various sentiments that are very dynamic, one example is the Simpel Pol Health Test application made by PT Cipta Sari Arsonia (CSA). The research objective is to obtain useful information from accurate community review sentiments for service improvement and feedback for service providers and application developers. The method used is Naïve Bayes Classifier with Tf-idf weighting, Multinomial Naïve Bayes with review value indicators and review sentences predicted by the BERT method as a determinant of sentiment value whether positive or negative. Sentiment towards the application shows quite encouraging results, from 3000 data analyzed with 1772 positive reviews and 263 negative reviews with 80% training data and 20% test data, the naïve bayes classification model is able to provide a high level of accuracy, which is 88.7% with a precision of 88.5%, recall of 100% and f1-score of 93.9%. The data showed that most people gave a positive response to this application, with the dominant word being 'easy'. This system was developed using the local-based streamlit framework and proved to be quite reliable in developing systems for data processing and web-based data analysis even though the scraping process is slightly longer than the google colab service. Future research is expected to be able to predict data that is positive or negative with several parameters and several sentiment analysis methods at once and their comparison.