Algorithmic agenda-setting: the subtle effects of news recommender systems on political agendas in the Danish 2022 general election

Abstract

ABSTRACT As news media strive to enhance their data science capabilities to compete with digital platforms, news recommender systems (NRSs) provide viable solutions for enhancing relevance, engagement, and cost efficiency. However, emulating the filtering practices of digital platforms raises concerns about democratic implications, including audience fragmentation and diminished editorial control. This study contributes to the existing literature on algorithmic news filtering through a controlled online experiment conducted on a Danish tabloid news platform during the Danish 2022 general election. The experiment involved a treatment group (n = 63.721) exposed to personalized recommendations with a non-exposed control group (n = 16.051). From an agenda-setting perspective, this study examines the exposure of political news, the salience and attributes of political actors and issues, and the diversity of the news composition. The main finding of this study is that the exposure effects of news recommender systems are significant but small. The effects are further not exclusively a direct result of the re-ranking of content but also indirect effects of changes in consumption patterns. Also, the study finds that personalized recommendations resulted in exposure to softer news content and a less diverse variation in the exposure of topics, actors, or issues. However, small effects can have substantial impacts, for instance, by systematic overexposure of certain types of political actors. Such effects are unanticipated and hard to discover, which stipulates the need to further the development of methods and practices for monitoring news algorithmic news curation.