The Structure and Dynamics of Diffusive Conversation

• In Event: Text as Data: Topic Models and Conversations

Abstract

Traditional models of communication fall into one of two distinct categories. \textit{Conversation} involves repeated, reciprocated interactions within a closed group, whereas \textit{diffusion} consists of the directed, one-off transmission of information in an unbounded environment. We argue that most human communication---from in-person speech to online social media posts---contains elements of both. We describe and validate a generalized statistical model of social interaction that nests conversation and diffusion as extreme cases, then develop techniques to address the computational demands of analyzing this communication at scale. Using anonymized email metadata from a large multinational corporation, we demonstrate that it is precisely this hybrid nature of email communication that enables social search and knowledge discovery within organizations. The model is then applied to a corpus of local government email communication in proximity to mayoral elections.

Author

• Dean Knox, Princeton University