User Demographics and Language in an Implicit Social Network

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1478–1488, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics User Demographics and Language in an Implicit Social Network Katja Filippova Google Inc. Brandschenkestr. 110 Z¨urich, 8004 Switzerland katjaf@google.com Abstract We consider the task of predicting the gender of the YouTube1 users and contrast two infor- mation sources: the comments they leave and the social environment induced from the af- filiation graph of users and videos. We prop- agate gender information through the videos and show that a user’s gender can be predicted from her social environment with the accuracy above 90%. We also show that the gender can be predicted from language alone (89%). A surprising result of our study is that the latter predictions correlate more strongly with the gender predominant in the user’s environment than with the sex of the person as reported in the profile. We also investigate how the two views (linguistic and social) can be combined and analyse how prediction accuracy changes over different age groups. 1 Introduction Over the past decade the web has become more and more social. The number of people having an iden- tity on one of the Internet social networks (Face- book2, Google+3, Twitter4, etc.) has been steadily growing, many users communicate online on a daily basis. Their interactions open new possibilities for social sciences, and linguistics is no exception. For example, with the development and growth of Web 2.0, it has become possible to get access to masses of text data labeled with respect to different social 1www.youtube.com 2www.facebook.com 3www.plus.google.com 4www.twitter.com parameters such as country, age, gender, profession or religion. The study of language varieties between groups separated by a certain social variable belongs to the field of sociolinguistics which more generally investigates the effect of society on how language is used (Coulmas, 1998). Historically, sociolinguistics is connected to dialectology whose focus has been primarily on the phonetic aspect of the regional di- alects but was later extended to sociolects (Cham- bers & Trudgill, 1998). A usual study would involve sampling speakers from a population, interviewing them and analyzing the linguistic items with respect to social variables (Hudson, 1980). The last decade has seen several studies inves- tigating the relationship between the language and the demographics of the users of blogs or Twitter (see Sec. 2 for references). Most of those studies used social network sites to collect labeled data– samples of text together with the demographics vari- able. However, they did not analyse how social en- vironment affects language, although very similar questions have been recently posed (but not yet an- swered) by Ellist (2009). In our work we attempt to address precisely this issue. In particular, we con- sider the task of user gender prediction on YouTube and contrast two information sources: (1) the com- ments written by the user and (2) her social neigh- borhood as defined by the bipartite user-video graph. We use …

Content truncated to 500 words