Klasifikasi Sentiment Review Aplikasi MyPertamina Menggunakan Word Embedding FastText dan SVM (Support Vector Machine)

Abstract

The MyPertamina application is a requirement for buying subsidized fuel oil (BBM), namely pertalite and diesel, the goal is that subsidized (BBM) purchases are right on target. The MyPertamina application has received many ratings and comments from the public, both positive and negative, with these comments and ratings expected to help the government as a benchmark in implementing a program. Therefore, this research aims to assess the MyPertamina application by grouping sentiment classes 90:10, 80:20 and 70:30. In this study, the method used is Fasttext and Support Vector Machine (SVM) to review the MyPertamina application. This research uses 8000 data, the data is grouped into three portions of data, with portions of 90:10, 80:20 and 70:30. The best SVM model was obtained with a data portion of 90:10 with a total of 7200 training data and 800 testing data, obtained 80% accuracy, 50% recall and 84% precision without undersampling. Meanwhile, if the amount of data is balanced (undersampling) with the number of positive data 1325, neutral 1325 and negative 1325, that is, with the benchmark of the lowest data value from the sentiment class, an accuracy of 67% is obtained, recall is 69% and precision is 57%. The highest number of sentiment classes from the 90:10 portion of the data is negative, namely 4300, neutral 1575 and positive 1325, because many users found reviews of the MyPertamina application, namely "after updating the MyPertamina application the bugs are getting worse".