Comparison of Machine Learning Methods in Sentiment Analysis PeduliLindungi Applications
Widya Cholil, Febriyanti Panjaitan, Ferdiansyah Ferdiansyah, Artika Arista, Ria Astriratma, Tri Rahayu
Abstract
KOMINFO, in partnership with the Ministry of State-Owned Enterprises (BUMN), developed the “PeduliLindungi” smartphone application, which has been implemented in the community. This application is unquestionably a topic that is frequently debated by the community via social media, namely Twitter. This PeduliLindung program harvests the positives and negatives of Twitter public opinion. Through posting on Twitter and expressing opinions on this problem, data can be collected and evaluated into negative and favorable opinion sentiments. From this occurrence, study was undertaken to analyze Twitter users' perspectives on the application's security. A dataset including 1,922 tweets will be analyzed using five Machine Learning techniques: Naive Base, SVM, Decision Tree (DT), K-Nearest Neighbor (KNN), and Random Forest. Then, it will be compared to the value of accuracy and AUC score for the performance of each of the five approaches. Collecting user tweet data with crawling techniques, data processing, word weighting with the TD-IDF approach, modeling with five machine learning methods, comparison, and algorithm selection are the stages of the suggested study method. The results indicated that the accuracy of the KNN method was 98.40%, Decision Tree 97.60%, Random Forest 97.60%, SVM 97.10%, and Naive Bayes 87.70%, while the AUC score for each method was 99.80% for KNN, 99.80% for Decision Tree, 99.80% for Random Forest, 95.80% for SVM, and 91.27% for Naive Bayes. With an accuracy value of 98.40% and an AUC Score of 98.80%, the KNN approach gives superior performance than the other methods.