Using topic models to measure social psychological characteristics in online social media

Abstract

Despite a growing body of research into computational models of social psychological processes, direct empirical grounding for these models remains an elusive goal. This is largely due to the difficulty of measuring modelled characteristics of social groups. This paper presents a methodology combining supervised topic models with traditional psycho-linguistic research as a first step towards such a goal. The method is applied to a collection of over a million tweets from the Twitter ‘pro-anorexia’ community.