Inferring User Social Class in Online Social Networks

Renato Miranda Filho, Guilherme R. Borges, Jussara M. Almeida, Gisele L. Pappa

Abstract: Although information posted in online social networks has proven to be accurate enough to monitor and predict real world phenomena, not a lot is known about users spreading this information. Previous studies have explored user public data to infer personal attributes such as gender, age and location, but one aspect is yet to be explored: social class. Assuming an objective definition of social class, based on income and wealth, we propose a new method to automatically generate a user social class dataset, taking advantage of Foursquare user interactions and Twitter messages. The basic idea to build our social class dataset is: the wealthier the place, the richer the users who usually visit it. We build our dataset by describing users using the contents of their tweets, and a machine learning algorithm is employed to automatically generate social class classification models. Our experimental results show that, considering social class divisions into two, three and four segments, the predictive accuracies of our models varied from 57% to 73%.