ABSTRACT
To stop the spread of the COVID-19, the Indonesian government implemented community activities restrictions enforcement (in Indonesian language: Pemberlakuan Pembatasan Kegiatan Masyarakat or PPKM) starting from January 2021. The term PPKM applied are PPKM Mikro (in Indonesian language) or Micro PPKM, PPKM Darurat (in Indonesian language) or Emergency PPKM, and PPKM Level 1-4 or Level 1-4 PPKM. On the other hand, the existing research mostly used Twitter as the data source to do sentiment classification. Therefore, we aimed to classify social media comments on Facebook and YouTube on Level 1-4 PPKM policy in Jakarta. We used "PPKM Jakarta"as the keyword topic in August - September 2021 when Level 1-4 PPKM was ongoing. In addition, we compared datasets composition, machine learning models, and features extraction. Random Forest, Naive Bayes, and Logistic Regression were performed as the machine learning models due to they were the top three models on the previous research. We extracted word unigram, word bigram, character trigram, and character quadrigram as the feature extraction. The highest average F-measure was obtained with a 79.6% score of the Logistic Regression model using character quadrigram extraction. We found that comments from Facebook and YouTube were dominated by neutral sentiment (49.8%) with this setup. It means the people of Jakarta started to trust the government in handling the COVID-19 pandemic. Through word cloud analysis, it is recommended that social assistance be reviewed for those directly affected. © 2021 IEEE.