The Use of Supervised Learning Algorithms in Political Communication and Media Studies: Locating Frames in the Press

Abstract

To locate media frames is one of the biggest challenges facing academics in Political Communication disciplines. The traditional approach to the problem is the use of different coders and their subsequent comparison, either through statistical analysis, or through agreements between them. In both cases, problems arise due to the difficulty of defining exactly where the frame is as well as its meaning and implications. And, above all, it is a complex process that makes it very difficult to work with large data sets. The authors, however, propose the use of information cataloging algorithms as a way to solve these problems. These algorithms (Support Vector Machines, Random Forest, CNN, etc.) come from disciplines linked to neural networks and have become an industry standard devoted to the treatment of non-numerical information and natural language processing. In addition, when supervised, they can be trained to find the information that the researcher considers pertinent. The authors present one case study, the media framing of the refugee crisis in Europe (in 2015) as an example. In that regard, SVM shows a lot of potential, being able to locate frames successfully albeit with some limitations.