In media discourses about topics like climate change, population aging, or the financial crisis, interest groups and politicians frequently voice their arguments or policy preferences. These statements of political actors about normative concepts can be analyzed using social network analysis in order to identify coalitions of actors, explain major reforms, or infer the underlying social and political dynamics. However, annotating statements of actors manually (e.g., in the software Discourse Network Analyzer) is very time-consuming. This research project tries to improve the situation by implementing and testing supervised (semi-automatic) machine learning algorithms. The prediction algorithms propose new actor-concept statements based on previous manual annotation. Beside the implementation of these methods, the major goal of the project is to assess the degree of expert knowledge which is necessary for a sufficiently valid semi-automatic identification of statements in new text portions.
