Characterizing the language-production dynamics of social media users

In this paper, we propose a characterization of social media users based on language usage over time in order to make more rigorous the notions of organic and inorganic online behavior. This characterization describes the extent to which a user's word usage within a particular time period subverts expectations based on preceding time periods. To do this, we adapt the use of an information theoretic measure of cognitive surprise and apply it to a set of behaviorally diverse Twitter users. We then compare the language-production dynamics across users based on term frequencies at multiple levels of granularity. We then illustrate the intuition behind this characterization through case studies of salient users identified from this method. Through these case studies, we find that this characterization can be linked to the degree to which a user's word usage is organic, inorganic, or a mixture of both.