Vrijenhoek, a computer scientist by training, conducted her doctoral research at the Institute for Information Law (IViR) at the University of Amsterdam. Her dissertation, titled ‘Building normative diversity into algorithmic news recommendations’, will be defended on 1 July at the Faculty of Law. The central question of her research is how abstract journalistic values, such as diversity, can be translated into concrete technical implementations within recommendation systems.

Personalised news recommendations extend well beyond social media, Vrijenhoek emphasises. Any system with access to large amounts of data that selects and ranks content falls within this category. On news websites, this is visible in sections such as “Read also”, where similar articles are suggested based on prior reading behaviour. Fully personalised news offerings remain rare in practice: editorial teams are reluctant to hand their values and standards over to an automated system, partly because such systems are still unable to adequately replicate journalistic judgement.

The risk of recommendations based purely on user behaviour is that readers are served exclusively entertainment-oriented content, while relevant but less click-friendly information disappears. Vrijenhoek therefore does not advocate replacing editors, but calls for technology designed with journalistic goals as its foundation. She also notes that aiming for a perfect solution from the outset can be paralysing: starting small with something comprehensible and verifiable is the more practical approach.

News personalisation does not necessarily lead to a filter bubble, according to Vrijenhoek. Technology can equally be configured to expose readers to viewpoints they have not yet encountered. She gives the example of a section titled “Political positions you do not yet know”, or a system that gives all political parties equal coverage. At the same time, there is also a case for a system that gives more prominent parties more attention. The right choice differs per news organisation. Vrijenhoek stresses that editorial teams must first determine what a diverse news offering means to them before technical specialists can begin implementing it.