Automated Content Analysis of Online Political Communication

Abstract

Content analysis has a long tradition in the social sciences: it is central to the study of policy preferences, propaganda and mass media, and the framing of social movements. New computational tools and the increasing availability of digitized documents promise to push forward this line of inquiry by reducing the costs of manual annotation and enabling the analysis of large-scale corpora. In particular, the automated analysis of online political communication may yield insights into political sentiment which offline analysis instruments (such as opinion polls) fail to capture; for instance, we are now in a better position to analyze the temporal dimension of opinion formation because of higher resolution data. Several linguistic peculiarities, however, distinguish online political communication from traditional political texts; for starters, it is less formal and structured. Automated content analysis techniques are also not always as reliable or as valid as manual annotation, which makes measurements potentially noisy or misleading. We provide an overview of techniques suited to two common content analysis tasks: classifying documents into specified categories, and discovering unknown categories from documents. This second task is more exploratory in nature: it helps identify topic domains when there are no clear preconceptions of the topics that are discussed in a certain communication environment; the first task, on the other hand, can help label a large volume of text in a more efficient manner than manual annotation. This chapter focuses on the application of these automated techniques to online political communication, and suggests directions for future research in this domain.