Using structural topic modeling to detect events and cluster twitter users in the Ukrainian Crisis

Abstract

Structural topic modeling (STM) is a recently introduced technique to model how the content of a collection of documents changes as a function of variables such as author identity or time of writing. We present two proof-of-concept applications of STM using Russian social media data. In our first study, we model how topics change over time, showing that STM can be used to detect significant events such as the downing of Malaysia Air Flight 17. In our second study, we model how topical content varies across a set of authors, showing that STM can be used to cluster Twitter users who are sympathetic to Ukraine versus Russia as well as to cluster accounts that are suspected to belong to the same individual (so-called “sockpuppets”). Structural topic modeling shows promise as a tool for analyzing social media data, a domain that has been largely ignored in the topic modeling literature.