Comparison of Kernel Support Vector Machine Multi-Class in PPKM Sentiment Analysis on Twitter

Abstract

PPKM is the Indonesian government's policy to deal with the spread of the coronavirus since early 2021. Until now, PPKM is still the main topic to prevent the spread of COVID-19. This policy has generated various responses from the public, especially on Twitter. A sentiment analysis process is needed to process the text obtained from Twitter. Sentiment analysis is a form of representation of text mining and text processing. This study aims to analyze public sentiment towards PPKM through data obtained from Twitter using the multi-class SVM algorithm. In implementing multi-class SVM, an analysis of the Polynomial and RBF kernels was carried out on the One Against One and One Against Rest methods which showed that the combination of One Against Rest and the Polynomial kernel obtained the best accuracy, which was 98.9%. Unlike the case with the combination of One Against One and Kernel RBF, which obtained the worst accuracy, 77.6%. The best model produces precision, recall, and f1-score values ​​of 97%, 98%, and 97%. Based on the confusion matrix results, the best model has a positive class distribution = 912, neutral = 51, and negative = 26. Overall, the model that uses polynomial kernel produces higher accuracy, both applied to the One Against One and One Against Rest methods. In contrast, the model that uses the RBF kernel produces lower accuracy and is significantly different when applied to the One Against One and One Against Rest methods. The model results show that public sentiment towards the PPKM policy is positive to be continued consistently to suppress the spread of the COVID-19 virus.