Eike Mark Rinke, Cäcilia Zirn, Charlotte Löb, Hartmut Wessler
Combining expert surveys and topic modeling to identify unknown issues in multinational media discourses

67th Annual Conference of the International Communication Association, San Diego, CA, May 26th to May 29th, 2017

Topic models allow for the discovery of previously unknown topics in news discourse. However, in contexts unfamiliar to the researcher this benefit is often difficult to realize because the researcher lacks the domain knowledge needed to gain relevant insights from the clustering solutions. In this paper, we present a general, applied solution to this common problem. The approach leverages deep domain knowledge obtained through national expert surveys to produce insightful topic model solutions. We illustrate it in a case study combining national expert surveys in six countries (Australia, Germany, Lebanon, Switzerland, Turkey, and the US) with a massive dataset of content from political blogs and online newspapers. The approach reveals cross-national variation in the issues discussed under a shared “master debate” (the public role of religion in societal life) as well as how “born digital” media contribute issues to this debate that are neglected in legacy online media.