An analysis of the user occupational class through Twitter content

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 1754–1764, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics An analysis of the user occupational class through Twitter content Daniel Preot¸iuc-Pietro1, Vasileios Lampos2 and Nikolaos Aletras2 1 Computer & Information Science, University of Pennsylvania 2 Department of Computer Science, University College London danielpr@sas.upenn.edu, {v.lampos,n.aletras}@ucl.ac.uk Abstract Social media content can be used as a complementary source to the traditional methods for extracting and studying col- lective social attributes. This study focuses on the prediction of the occupational class for a public user profile. Our analysis is conducted on a new annotated corpus of Twitter users, their respective job titles, posted textual content and platform-related attributes. We frame our task as classifi- cation using latent feature representations such as word clusters and embeddings. The employed linear and, especially, non-linear methods can predict a user’s occupational class with strong accuracy for the coars- est level of a standard occupation taxon- omy which includes nine classes. Com- bined with a qualitative assessment, the derived results confirm the feasibility of our approach in inferring a new user at- tribute that can be embedded in a multitude of downstream applications. 1 Introduction The growth of online social networks provides the opportunity to analyse user text in a broader context (Tumasjan et al., 2010; Bollen et al., 2011; Lam- pos and Cristianini, 2012). This includes the social network (Sadilek et al., 2012), spatio-temporal in- formation (Lampos and Cristianini, 2010) and per- sonal attributes (Al Zamal et al., 2012). Previous research has analysed language differences in user attributes like location (Cheng et al., 2010), gender (Burger et al., 2011), impact (Lampos et al., 2014) and age (Rao et al., 2010), showing that language use is influenced by them. Therefore, user text al- lows us to infer these properties. This user profiling is important not only for sociolinguistic studies, but also for other applications: recommender systems to provide targeted advertising, analysts who study different opinions in each social class or integra- tion in text regression tasks such as voting intention (Lampos et al., 2013). Social status reflected through a person’s occu- pation is a factor which influences language use (Bernstein, 1960; Bernstein, 2003; Labov, 2006). Therefore, our hypothesis is that language use in social media can be indicative of a user’s occu- pational class. For example, executives may write more frequently about business or financial news, while people in manufacturing positions could re- fer more to their personal interests and less to job related activities. Similarly, we expect some cate- gories of people, like those working in sales and customer services, to be more social or to use more informal language. Focusing on the microblogging platform of Twit- ter, we explore our hypothesis by studying the task of predicting a user’s occupational class given platform-related attributes and generated content, i.e. tweets. That has direct applicability in a broad range of areas from sociological studies, …

Content truncated to 500 words