Using External Information for Classifying Tweets

Automatic classification of texts by topic is a well-studied problem. Nonetheless, classifying twitter messages by topic is difficult because the messages are short and the features space for classification is very sparse. We propose a method to enhance the text of the messages that contain links with external information such as the title of the web pages, and with the most frequent terms from these web pages. We show that the results of the classification improve substantially when adding this external information.