RADio-: a simplified codebase for evaluating normative diversity in recommender systems

Diversity is one of the core beyond-accuracy objectives considered in the development of news recommender systems. However, there is a clear gap between its technical conceptualization, typically as an intra-list distance, and a more normative interpretation, which touches upon the role the recommender system plays in society. Vrijenhoek et al. [1] proposed to instead use rank-aware divergence metrics to express normative diversity in news recommendations. This work describes a repository that allows for easy implementation of these metrics, by making the different diversity aspects and tactics configurable. It also contains an example implementation and analysis of the results. With its modular setup, the repository thus allows for conceptualizations of diversity that can be tailored to the news domain they need to be applied in.