A Semi-automated Pipeline for Mapping the Shifts and Continuities in Media Discourse

Mass media are an important actor between the authorities and citizens. Mass media frame the news and set the agenda for public debates. Investigation of this role of media has inspired various techniques for the analysis of media discourse in social and political research. However, manual coding of large corpora takes a long time and is prone to bias. Topic modeling can automate the search for key terms but it can hardly trace the shifts in a debate if the key terms are changing over time. In this paper, we propose a clustering-based pipeline that automates the search of key terms in the media discourse on a given topic and traces their development in time. The proposed technique helps to connect the clusters of unique terms across time periods into ‘discursive streams’. Such streams’ relative proportions and contents can be compared on a resulting map, a river network. The steps for creating such maps are explained. Two use cases demonstrate the shifts and continuities in the national media discourse about Internet regulation and labor migration in Russia over almost a decade. The analyses are based on more than 5,000 texts coming from the Integrum and Public.ru media archives. The resulting networks show which discussions evolved in the media discourse in the media coverage of these topics; how the key terms changed over time within the same debates; how the discursive streams grew in size and died out. The results are discussed in terms of validity and the applicability of the proposed technique to the study of media coverage of other topics.