The master thesis was written by Gabriel Grill.
Abstract
The quantitative study of news media can provide insights into reporting patterns and enable public discourse. Print media is essential in democratic societies, so its study remains important. This thesis examines Austrian reporting using network-based methods and unpacks the suitability of such an approach for news analysis. We seek to contribute to scholarly debates around the benefits and risks of using such methods to make sense of reporting. We conduct a comparative analysis of six news outlets (Der Standard, Die Presse, Österreich, Die Kronen Zeitung, Kurier, Die Heute) and discuss differences in reporting during the 2016 Austrian presidential elections. This is the first research effort applying a network-based approach to the Austrian Media Corpus (AMC), a complete and unique collection encompassing the last three decades of Austrian media coverage. We use natural language processing to extract an expressive subset of named entities representing network nodes. Several entity recognition schemes are evaluated based on a set of labeled articles. An approach combining named entity linking based on a Wikidata dictionary with an open-source recognition model performed best. The networks are constructed by relating the extracted entities and certain terms when they co-occur in a sentence. We apply various algorithms to the resulting networks to rank entities according to relevance and cluster nodes to detect themes during the election. We provide several descriptive statistics on publishing patterns and the occurrence of entities in the newspapers, revealing gender bias and the presidential candidates most mentioned. Our network-based approach reveals differences in reporting compared to results based on counting mentions. However, we also point out issues of these methods, such as flexibility in parameterization and messy visualizations. We argue these challenges are a double-edged sword as, for example, flexibility may also give researchers agency to enable more exploration and qualitative interpretation.
Supervision
Supervisor: Hannes Werthner, Co-Supervisor: Julia Neidhardt
