Individual Submission Summary
Share...

Direct link:

Download

A Three-country Analysis of Online Political Fragmentation Using Topic Modelling

Fri, August 30, 10:00 to 11:30am, Hilton, Embassy

Abstract

Political fragmentation in the digital domain has been at the center of academic research for more than a decade. In this paper, we investigate how topic overlap between media outlets and political actors can be used to measure levels of fragmentation across lines of political difference in the digital domain. To this end, we use a structural topic model (STM) and analyze over 70 million tweets collected during the 2017 national elections in France, Germany and the UK.

Previous research has measured fragmentation by looking at the extent to which public opinion shares common ground (Lee, 2007; McCombs, 2005; Sunstein, 2001), how audiences navigate the online news domain (Fletcher & Nielsen, 2017; Majo-Vazquez, Nielsen, & Gonzalez-Bailon, 2018; Webster & Ksiazek, 2012) or more generally, the provision of information on the web (Chaffee & Metzger, 2001; Hindman, 2018; Pariser, 2011). Yet, there still exist two competing hypotheses on whether and to what extent the online domain is fragmented. In this paper we aim at contributing to this unresolved and relevant debate by assessing the relationship between political parties and media outlets in articulating online electoral discourses ahead of three of the most important elections in Europe in 2017. Our research goal is to measure the fragmentation in the public sphere by assessing topic overlap between news media and political parties. In other words, we analyze similarities and differences between the media and political narratives during these major political events. Even more relevant, we focus here on whether patterns of topic overlap vary across lines of political difference. Are right-wing political parties and news outlets connected in the online domain by the amount of shared political topics? Do they jointly boost the prominence of similar topics? Are news outlets and parties at the extreme of the ideological spectrum more closely related to each other? Are divisive issues like immigration, religion or refugees, which were central during the three electoral contests, boosted by news outlets and political parties that share similar ideological positions?

To answer our research questions, we first show the robustness of our choice of topics by means of endogenous measures (semantic coherence, exclusivity) and external validations (topic relations, stability of semantic space). Then, we investigate the extent to which politicians and outlets talk about such topics over time (i.e. topic ownership). Further, STM allows us to determine how different political actors and news outlets use different words to express the same topics (i.e. topic alignment).

The contribution of the analysis presented in this study is two-fold. On one hand, it adds to our understanding of the underlying mechanisms behind the current level of fragmentation in the online domain. By investigating the topic alignment between news outlets and political parties and the extent to which the electoral debates were dominated by core topics, we contribute novel knowledge to the studies looking at the structure of the online audience attention. On the other hand, the events we examine in this study were particularly interesting because of the role played by the populist and extreme right-wing political parties like the AfD in Germany and Front National in France. Our methodological approach allows us to identify the topics that these parties were more interested in and whether they aligned with certain types of news media outlets i.e., digital-born or tabloids in boosting the prominence of those topics online. To do so, we first, assess levels of topic alignment between actors and then, we calculate the ideological position of the media outlets based on data from the European Election Study Survey and the Digital News Report (Newman, Fletcher, Kalogeropoulos, Levy, & Nielsen, 2018). Finally, we borrow tools from network science to run a community analyses that identify groups of news outlets and political actors that were more interested in the same topics across the election campaign.

Overall, in this study we map the semantic space of political actors and media outlets across countries. And hence, we contribute to the current debate on online fragmentation providing a comparative perspective that is missed in most of the above-mentioned studies (the exception being Fletcher & Nielsen, 2017; Majo-Vazquez et al., 2018). Preliminary results for the UK show that politicians and media outlets tend to tweet about the same topics, yet they differ in the words they use. Specifically, they refer to politicians they are ideologically close to, and express concerns that are consistent with their own beliefs.

Authors