Using Conceptual Class Attributes to Characterize Social Media Users

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 710–720, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Using Conceptual Class Attributes to Characterize Social Media Users Shane Bergsma and Benjamin Van Durme Department of Computer Science and Human Language Technology Center of Excellence Johns Hopkins University Baltimore, MD 21218, USA Abstract We describe a novel approach for automat- ically predicting the hidden demographic properties of social media users. Building on prior work in common-sense knowl- edge acquisition from third-person text, we first learn the distinguishing attributes of certain classes of people. For exam- ple, we learn that people in the Female class tend to have maiden names and en- gagement rings. We then show that this knowledge can be used in the analysis of first-person communication; knowledge of distinguishing attributes allows us to both classify users and to bootstrap new train- ing examples. Our novel approach enables substantial improvements on the widely- studied task of user gender prediction, ob- taining a 20% relative error reduction over the current state-of-the-art. 1 Introduction There has been growing interest in characteriz- ing social media users based on the content they generate; that is, automatically labeling users with demographic categories such as age and gender (Burger and Henderson, 2006; Schler et al., 2006; Rao et al., 2010; Mukherjee and Liu, 2010; Pen- nacchiotti and Popescu, 2011; Burger et al., 2011; Van Durme, 2012). Automatic user character- ization has applications in targeted advertising and personalization, and could also lead to finer- grained assessment of public opinion (O’Connor et al., 2010) and health (Paul and Dredze, 2011). Consider the following tweet and suppose we wish to predict the user’s gender: Dirac was one of my boyhood heroes. I’m glad I met him once. RT Paul Dirac image by artist Eric Handy: http:... State-of-the-art approaches cast this problem as a classification task and train classifiers using super- vised learning (Section 2). The features of the classifier are indicators of specific words in the user-generated text. While a human would as- sume that someone with boyhood heroes is male, a standard classifier has no way of exploiting such knowledge unless the phrase occurs in training data. We present an algorithm that improves user characterization by collecting and exploiting such common-sense knowledge. Our work is inspired by algorithms that pro- cesses large text corpora in order to discover the attributes of semantic classes, e.g. (Berland and Charniak, 1999; Schubert, 2002; Almuhareb and Poesio, 2004; Tokunaga et al., 2005; Girju et al., 2006; Pas¸ca and Van Durme, 2008; Alfonseca et al., 2010). We learn the distinguishing attributes of different demographic groups (Section 3), and then automatically assign users to these groups whenever they refer to a distinguishing attribute in their writings (Section 4). Our approach obviates the need for expensive annotation efforts, and al- lows us to rapidly bootstrap training data for new classification tasks. We validate our approach by advancing the state-of-the-art on the most well-studied user clas- sification task: predicting user …

Content truncated to 500 words