Bank Syariah Indonesia (BSI) is an active topic of conversation on Twitter, but customer sentiment patterns towards the bank's services have not been quantitatively analyzed. This study performs positive and negative sentiment classification on 24,401 Indonesian tweets collected on May 17, 2023. The preprocessing stage includes text cleaning, nonstandard word normalization, stopword removal, and stemming with the Sastrawi library. The data was labeled based on the affection dictionary and verified manually. Text representation is done with word frequency-based unigram-bigram method using CountVectorizer, then trained using Multinomial Naive Bayes algorithm. Evaluation of the model against test data resulted in an accuracy of 94%, with precision, recall, and F1-score of 93% each. Words that commonly appear in positive sentiments include easy and fast service, while negative sentiments are dominated by the words error and maintenance. These results show that the Naive Bayes-based approach and word frequency representation are effective for rapid analysis of public opinion towards BSI on social media.