The dataset was created from 5000 tweets that were manually classified by 5 people. If discrepancies were detected, a majority voting scheme was used in order to tag the tweets' intentions. In order to perform the experimental validation of the automatic classification with the machine learning algorithms, several text representations were used (tf-idf, tf-rfl and bin-rfl). The results obtained from the various tests that were conducted yielded satisfactory results. We also analyzed how to assign automatic responses to frequently asked questions, and obtained promising results.