Rethinking news algorithms: nudging users towards diverse news exposure
Can we design news recommenders to nudge users towards diverse consumption of topics and perspectives? The growing role of news recommenders raises the question of how news diversity can be safeguarded in a digital news landscape. Many existing studies look at either the supply diversity of recommendations, or the effects of (decreased) exposure diversity on e.g. polarization and filter bubbles. Research on how users choose from the available supply is lacking, making it difficult to understand the relation between algorithm design and possible adverse effects on citizens.
We directly address the question of how news recommender algorithms can be designed to optimize exposure diversity. This innovative and interdisciplinary project builds on our extensive expertise on news diversity, news consumption behavior, text analysis and recommender design by providing: (WP1) a normative framework for assessing diversity; (WP2) a language model to automatically measure diversity; (WP3) a model of news consumption choices based on supply, presentation, and individual characteristics; and (WP4) a concrete prototype implementation of a recommender algorithm that optimizes exposure diversity, which will be externally validated in a unique field experiment with our media partners.
The project will bridge the gap between differing understandings of news diversity in computer science, communication science, and media law. This will increase our understanding of contemporary news behavior, yield new language models for identifying topics and perspectives, and offer concrete considerations for designing recommenders that optimize exposure diversity. Together with media companies and regulators we turn these scientific insights into concrete recommendations.