Indonesia is one of the five countries with the most Twitter users in the world, as many as 77% are active users. Twitter makes it easy for people to convey opinions or information through tweets. The topics discussed by the public on Twitter vary, one of which is the Certified Internship program which is one of the Certified Independent Study Internship programs (MSIB) where this program was designed by the Ministry of Education, Culture, Research and Technology with the aim of providing opportunities for students to learn and develop yourself in the world of professional industry for 1-2 semesters. This program raises sentiments in the community so that sentiment needs to be analysed, so that information from sentiments on Twitter tweets can be useful by classifying sentiments into positive, neutral and negative. In this study, data was obtained from web scarping using the Python programming language. The data obtained from scraping was 14,586, then the data was preprocessed by going through the stages of cleansing, case folding, tokenizing, filtering slangwords, removing stopwords, to sentences, stemming and finally dropping duplicate and empty tweets. The results of the data preprocessing produce the final data of 11,919 data used for the labeling process. Data labeling was carried out using the Lexicon Based method by obtaining a total of 9,595 positive sentiments, 14 data of neutral sentiment and 2,310 negative sentiments. Then the Naïve Bayes classification algorithm is applied with the Naïve Bayes Multinomial model with a ratio of training and test data of 8:2. The test results carried out on certified apprentice tweets obtained an Acurracy Score of 85%, with the highest Precision score of 87%, the highest Recall of 95% and the highest F1-Score value of 91%.