Czech Coal Speakers – Extraction of Networks of Actors from Media Texts

It has been shown that media norms might affect and potentially bias the way how news stories are constructed – a case already well-documented in media research (cf. Boykoff and Boykoff 2007). The media tend to allow representatives of the establishment to present their position on the issue (Boykoff and Boykoff 2007; Boykoff 2011), allowing them to act as important speakers and thus shape the media discourse. This article is going to investigate this proposition through the implementation of the social network analysis on a substantive case of coal mining in the Czech Republic, the former Eastern Bloc member and the third largest coal consumer in the EU, through social network analysis of co-occurrence of individuals in the Czech quality press. Coal mining is an important energy and climate issue in the country, since it facilitates a 15-year old conflict over the existence of coal-restricting policies, implemented to instigate coal phase-out and to mitigate the degradation of the local environment (Rečka and Ščasný 2017). While countries like Germany attempt to free themselves from coal conundrum, the trend in the Czech Republic is not that straightforward. In this setting of a polarized and ongoing national conflict, media coverage plays an important role due to agenda-setting effects (McCombs 2004). We investigate, how the „speakers” on the issue co-occur in the media, based on the co-occurrence in individual media articles. Automated text mining is used as the main tool to parse and extract network data from existing media texts. This is a scalable and easy to employ an alternative to human coding of text content, and its application has been already tested and employed (cf. van Atteveldt, Kleinnijenhuis, and Ruigrok 2008). This article relies on named entity recognition for extraction of organizations and individuals, which appear in individual articles. Along with that, gender and age of individuals are obtained using through as part-of-speech tagging. This way, a list of basic attributes is constructed along with the identification of individuals in the network. Basic network metrics, such as centrality measures (Freeman 1978) would allow identifying the most important actors in the media discourse. The output of the tool is evaluated vis-à-vis manually coded sample of media articles, performed by the expert coder. Lastly, the analysis is paired both with article metadata, allowing to break the discourse down based on publisher and time of publication, as well as with unsupervised text categorization (Blei, Ng, and Jordan 2003) in order to provide more insight into the topics of media debates, allowing to further differentiate networks by article focus.