The Assumptions Built into (Research on) News Recommender Systems

The next session at Future of Journalism 2023 conference that I’m attending is on polarisation, so of course I had to check it out; it starts with Mel Bunce and her colleagues’ study of the Media Freedom Coalition. However, they’ve asked for this study not to be tweeted at this stage, so I shall also not blog about it for now.

The second speakers in this session, then, are Jannie Møller Hartley and Elisabetta Petrucci, whose interest is in diversity in news recommender systems. Such systems may involve content filtering (based on the content of news articles) and collaborative filtering (based on the usage patterns of similar users) aspects, and their operation can affect the diversity of news content that actual news users encounter. But what normative assumptions about journalism are involved in these concerns about diversity as they are discussed in the literature?

The present study examined the literature on news diversity and recommender systems, covering some 32 articles, and examined how these framed this issue. These largely framed diversity as a necessary precondition for the functioning of diversity, and this is often operationalised as content diversity (across news beats) or political viewpoint diversity (across ideologies). It is taken for granted that any problems with a lack of diversity can be solved by recommender systems that incorporate diversity by design.

Journalism is widely seen as allowing citizens to participate in democracy; the need to have recommender systems is seen as a given; and there is a fundamental assumption that exposure diversity as a result of diversity by design leads automatically to consumption diversity, too. But the latter automatism is unlikely to exist in practice; consumption within a single news site does not reflect the potentially more diverse news repertoire of individuals in general anyway; and journalism has a broader role than just to cover politics. It is necessary, therefore, to question these underlying assumptions much more strongly, then (as well as the broader assumption that the operation of recommender systems locks users in ‘filter bubbles’, anyway).

Indeed, there is a need to consider whether recommender systems are necessary at all, and to explore who builds these systems and what assumptions they incorporate into them. This also requires a more holistic perspective on individual news repertoires and the societal role of journalism in the first place.