Discriminating Gender on Twitter

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1301–1309, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Discriminating Gender on Twitter John D. Burger and John Henderson and George Kim and Guido Zarrella The MITRE Corporation 202 Burlington Road Bedford, Massachusetts, USA 01730 {john,jhndrsn,gkim,jzarrella}@mitre.org Abstract Accurate prediction of demographic attributes from social media and other informal online content is valuable for marketing, personalization, and legal in- vestigation. This paper describes the construction of a large, multilingual dataset labeled with gender, and investigates statistical models for determining the gender of uncharacterized Twitter users. We explore several different classifier types on this dataset. We show the degree to which classifier accuracy varies based on tweet volumes as well as when various kinds of profile metadata are included in the models. We also perform a large-scale human assessment us- ing Amazon Mechanical Turk. Our methods signifi- cantly out-perform both baseline models and almost all humans on the same task. 1 Introduction The rapid growth of social media in recent years, exem- plified by Facebook and Twitter, has led to a massive volume of user-generated informal text. This in turn has sparked a great deal of research interest in aspects of so- cial media, including automatically identifying latent de- mographic features of online users. Many latent features have been explored, but gender and age have generated great interest (Schler et al., 2006; Burger and Henderson, 2006; Argamon et al., 2007; Mukherjee and Liu, 2010; Rao et al., 2010). Accurate prediction of these features would be useful for marketing and personalization con- cerns, as well as for legal investigation. In this work, we investigate the development of high- performance classifiers for identifying the gender of Twitter users. We cast gender identification as the ob- vious binary classification problem, and explore the use of a number of text-based features. In Section 2, we de- scribe our Twitter corpus, and our methods for labeling a large subset of this data for gender. In Section 3 we discuss the features that are used in our classifiers. We describe our Experiments in Section 4, including our ex- ploration of several different classifier types. In Section 5 we present and analyze performance results, and discuss some directions for acquiring additional data by simple self-training techniques. Finally in Section 6 we summa- rize our findings, and describe extensions to the work that we are currently exploring. 2 Data Twitter is a social networking and micro-blogging plat- form whose users publish short messages or tweets. In late 2010, it was estimated that Twitter had 175 million registered users worldwide, producing 65 million tweets per day (Miller, 2010). Twitter is an attractive venue for research into social media because of its large vol- ume, diverse and multilingual population, and the gener- ous nature of its Terms of Service. This has led many re- searchers to build corpora of Twitter data (Petrovic et al., 2010; Eisenstein et al., 2010). In April 2009, we began sampling data …

Content truncated to 500 words