Automatically inferring user demographics in social networks is useful for both social science research and a range of downstream applications in marketing and politics. Our main hypothesis is that language use in social networks is indicative of user attributes. This talk presents recent work on inferring a new set of socioeconomic attributes, i.e. occupational class, income and socioeconomic class. We define a predictive task for each attribute where user-generated content is utilised to train supervised non-linear methods for classification and regression, i.e. Gaussian Processes. We show that our models achieve strong predictive accuracy in all of the three demographics while our analysis sheds light to factors that differentiate users between occupations, income level and socioeconomic classes.