User Type Classification of Tweets with Implications for Event Recognition

Twitter has become one of the foremost platforms for information sharing. Consequently, it is beneficial for the consumers of Twitter to know the origin of a tweet, as it affects how they view and interpret this information. In this paper, we classify tweets based on their origin, exploiting only the textual content of tweets . Specifically, using a rich, linguistic feature set and a supervised classifier framework, we classify tweets into two user types - organizations and individual persons. Our user type classifier achieves an 89% F1-score for identifying tweets that originate from organizations in English and an 87% F1-score for Spanish. We also demonstrate that classifying the user type of a tweet can improve downstream event recognition tasks. We analyze several schemes that exploit user type information to enhance Twitter event recognition and show that substantial improvements can be achieved by training separate models for different user types.