Perbandingan antara Multinomial Naïve Bayes dan Regresi Logistik pada Analisis Sentimen Ulasan Aplikasi Sayurbox

Abstract

Sayurbox needs to know the satisfaction of users regarding the applications they make to keep users' loyalty. One way to find out users satisfaction is to look at the users' reviews through the Google Play Store. Sentiment analysis is useful for managing these reviews and allows reviews to have certain categories such as positive or negative. The classification process can be done manually, but it will take a long time if the number of reviews that need to be classified is large. Therefore, a model is required to classify reviews automatically for saving time. The algorithms to be used in this study are Multinomial Naïve Bayes and logistic regression, and these algorithms will be combined with different kinds of n-gram (unigram, bigram, trigram, unibigram, unitrigram, bitrigram). The study was conducted to compare both algorithms and the six types of n- gram, select the best model combinations, and predict review sentiment from June 2022 to July 2022. The results have shown that the use of two different algorithms and several types of n-gram had a significant effect on the accuracy value, AUC, and f1-score. Interaction between algorithms and n-gram also had significant influence. The result shows that Multinomial Naïve Bayes with unibigram has been the best algorithm because it has shown the highest average of accuracy (91,02%), AUC (83,66%), and f1-score (94,38%). Multinomial Naïve Bayes with unibigram was used to predict reviews from June 2022 – July 2022. It has been predicted that there were 153 positive reviews and 37 negative reviews.