The government may use social media, such as Twitter, to socialize a policy or a program to society. We may predict whether a program is successful or not by analyzing the sentiment of societies towards such program or communities through their tweets. The latest program of Indonesia's government during the COVID-19 pandemic is to make people do social distancing. It is socialized using the hashtag of stay at home appeal (#dirumahaja). The objective of this study is to analyze the understanding of societies regarding this program through people's tweet. We compared two classification algorithms (Naive Bayes and Random Forest), using tokenization and unigram features to build classification model of tweet sentiment. The tweets that included some hashtags regarding social distancing program, were collected with 5101 tweets in total. The highest accuracy is obtained using the Random Forest algorithm and term weighting feature, which yielded 95.98%. From the model we found that the number of positive sentiments is greater than the negative sentiment. Which can be concluded that the societies are understand and agree to the social distancing program.