Data amount becomes rapidly increased in today's era. Data can be in form of text, picture, voice, and video. Social media is one factor of the data increase as everybody expresses, gives opinion, and even complains in social media. The first step is data collection used API twitter with each candidate names on Jakarta Governor Election. The collected data then became input for preprocessing step. The next step is extracted-each tweet's feature to be listed. The list of features were transformed into feature vector in binary form and transformed again used Tf-idf method. Dataset consists of two kinds of data, training and testing. Training was labeled manually. K-Fold Cross Validation is used to test algorithm performance. Based on the result of the test, accuracy obtained reached 74% in average with composition of training data and testing data by 90:10. Changed folding amount gave no impact to the accuracy level.