Diverse News Recommenders: from concept to implementation

News recommender systems, driven by data and machine learning, automatically filter the enormous amount of news that is available online, and use this to populate newsletters, personalized news apps or social media news feeds. (Personalised) news recommendation systems are widely seen as one of the most promising AI innovations in the media for the years to come (Nic et al., 2018) – a technology that offers the potential for entirely new ways of informing and engaging with the audience, and offering innovative, value-added media products (Bodo, 2019). The most prominent concern about existing news recommenders is that using them could mean missing out on challenging viewpoints and important ideas (Thurman et al., 2019), while only including items users would want to read (commonly referred to as a filter bubble). These concerns are also reflected in the literature and policy debate about filter bubbles and echo chambers and the potentially negative consequences not only for users, but also for social cohesion, a resilient digital society and, ultimately, democracy. Fears about creating filter bubbles and echo chambers, finally, are also one of the primary obstacles that discourage the media from embracing this powerful new technology (Bodo, 2019; Diakopoulos, 2019). Part of the solution to these concerns is value-driven news recommender design that formalises diversity as societal value (Bastian et al., 2021).

The goal of this workshop is to further advance the conceptualisation of diversity as a concept that has a grounding in social science, is implementable in a technical setting, and usable in practice. With a group of experts from different disciplines as well as from both academia and industry, we will think about the question of what makes a good and diverse news recommendation, and translate these notions into scalable technical solutions. This workshop will be considered a success when we come up with a set of metrics of diversity that are sound from both a social and a technical perspective, and that can be incorporated in recommender system design.