Predicting latent attributes of Twitter user by employing lexical features

The rapid growth of social media, especially Twitter in Indonesia, has produced a large amount of user generated texts in the form of tweets. Since Twitter only provides the name and location of its users, we develop a classification system that predicts latent attributes of Twitter user based on his tweets. Latent attribute is an attribute that is not stated directly. Our system predicts age and job attributes of Twitter users that use Indonesian language. Classification model is developed by employing lexical features and three learning algorithms (Naïve Bayes, SVM, and Random Forest). Based on the experimental results, it can be concluded that the SVM method produces the best accuracy for balanced data.