Fine-Grained Sentiment Analysis on PeduliLindungi Application Users with Multinomial Naive Bayes-SMOTE

The Key Social Disability Policy (PSBB) requires highly mobile people to use the PeduliLindungi application. One application has several reviews from positive and negative users. Review data can be labeled with two types of emotions: negative sentiment and positive sentiment. Fine-grained sentiment analysis is a type of sentiment analysis that can be used to identify user reactions. One method of sentiment analysis is Multinomial Naïve Bayes. In this research, we used Multinomial Naïve Bayes to perform fine-grained sentiment analysis for users of the PeduliLindungi application. The data used is from the Google Play store. The sentiment class labeling results for the PeduliLindungi review data resulted in 9021 reviews, including a total of 6244 negative reviews and 2777 positive reviews. This research uses a data-sharing model that divides 80% of training data and 20% of test data. Many data imbalances for the two sentiment classes can be overcome by using the SMOTE method. SMOTE has been shown to improve classification accuracy more effectively than non-SMOTE, as applying SMOTE has been shown to improve the performance of imbalanced data. The proper classification method used to classify PeduliLindungi's user ratings is Multinomial Naïve Bayes-SMOTE, which has the highest AUC value.