The abundance of online content in today’s high-choice information environment creates new possibilities, but also challenges for communication research. One of these challenges is detection of content relevant for answering specific research questions, which is a task increasingly delegated to automated content analysis. In our paper we discuss lessons learned from designing automated classifiers for tackling two computationally challenging tasks in the field of text analysis: 1) detection of political content; and 2) detection of populist radical right (PRR) content. To approach these tasks as part of analysis of the large volume of cross-platform textual data in German generated by tracking online behavior of German and Swiss internet users, we used three computational approaches: 1) dictionary-based classification; 2) supervised machine learning-based classification; 3) deep-learning-based classification. In the paper we share our observations concerning the performance of the three approaches for the two tasks as well as the procedures used for their validation and optimization. Our findings highlight the potential of combining classification approaches (e.g., supervised machine learning and deep learning for tackling classification of PRR content) together with the importance of comparative inquiries in the field of automated content analysis.
