At Aftenposten, personalisation solves more problems than it creates

By Christoph Schmitz


Oslo, Norway


By Karl Oskar Teien

Schibsted News Media

Oslo, Norway


The problem with most of today’s online newspapers is that their front pages are updated at a frequency that does not correspond with how most readers actually use the front page.

A reader who is online at 11:00 a.m. will see a completely different front page and different stories than a reader who is online at 3:00 p.m. With print newspapers this was not a problem. The newspaper was static for 24 hours, and all readers had access to the same information regardless of when they read the newspaper.

When users of today’s online newspapers risk receiving a completely different idea of what the day’s most important stories are, we may end up with a fundamental problem for society. At Aftenposten, we have worked on an editorially controlled algorithm over five years that we believe can be part of the solution.

Schibsted's Aftenposten has found the solution of providing readers with the most important stories of the day along with personalised content based on interests and behaviour.
Schibsted's Aftenposten has found the solution of providing readers with the most important stories of the day along with personalised content based on interests and behaviour.

If we look at the disparity in readers’ behaviour and interests on Aftenposten’s digital platforms, the challenge with a digital broadcasting model is evident.

Here are some examples:

  • More than one-fifth of our users visit Aftenposten every day of the week, while just under one-fifth visit only one day a week.
  • Half of the subscribers visit only once a day, while only one in 10 visits more than four times a day. On an average day, a significant portion will visit us for the first time in the afternoon.
  • Readers only consume a small portion of the stories we publish every day.
  • Additionally, users’ interests in particular topics vary significantly. For example, 18% are not interested in sports at all, while 8% love sports.

How do we ensure that all these different people with different behaviours see the same top stories? How do we avoid a situation where those that are not interested in football at all see a wall of stories about Erling Haaland when they visit our front page for the first time that day while missing out on other major news stories?

Optimised for male (55)

With a broadcasting model, there is also a great danger of optimising for the largest demographic and the sum of all readers. For the largest media houses in Norway, this group is a man in his 50s.

We can certainly split traffic numbers into different segments, but it is difficult to avoid optimising for the average reader when creating one product for all users. The consequence is that narrower topics intended for particular audiences are deprioritised on the front pages, reducing the diversity of content that we serve our end users.

The editorially controlled algorithm

At Aftenposten, we believe in a combination of manually selected articles and an algorithm-driven ranking of stories on the front page. The top five or six placements on the front page are manually selected based on carefully considered editing principles. This ensures that the reader at any time will have a good overview of both today’s important stories and the latest top stories.

Additionally, we have the option to manually pin certain stories further down the front page. The rest of the front page is based on an algorithm that allows the reader to easily catch up on important news stories they have missed, and stories that have been widely read by others. Meta-data, such as the editorially determined “news value” and “lifetime” of an article, are a critical ingredient in achieving this.

In addition, we are working to incorporate signals about what the individual reader is likely interested in. This will never be the only thing we optimise for, but it will be an important tool for giving narrower topics a chance on a crowded front page.

The purpose is to help us match these narrow articles to the readers who have a special interest in that particular topic. This way of personalising differs significantly from what most people are used to from Facebook and YouTube.

Miles away from Silicon Valley

In essence, our algorithms help us rank a list of articles under the manual top stories, all of which are subject to our rigorous journalistic principles. The algorithms we are exposed to on social media, on the other hand, are based on an almost infinite amount of content and optimised for maximum engagement and habit formation.

Worrying about echo chambers is much more justified when a YouTube algorithm chooses from 800 million videos to maximise user engagement than when it ranks 80 Aftenposten articles using an algorithm with editorial signals.

The delicate balance of front page content curation resulted in a specific algorithm being used in Schibsted newsrooms.
The delicate balance of front page content curation resulted in a specific algorithm being used in Schibsted newsrooms.

The rules we set for our algorithms are based on an ambition to create long-term value and time well spent for subscribers, not to maximise short-term clicks or total time spent. This means that personalisation is not the same as “more of what I like.” Personalisation simply means that my content feed is adapted to me as a user while what it is optimised for depends on the goals we have set.

For Aftenposten, several goals are important:

  • Everyone must get the most important stories. By using editorial meta-data in combination with data on what others are reading, we ensure everyone, regardless of how often they visit us, gets today’s most important articles and most-read stories.
  • Readers should get stories they are particularly interested in. By capturing which stories readers follow over time and which topics they are interested in, we avoid readers missing out on stories they have a special interest in. Most readers consume a relatively small selection of the total number of published stories and need help discovering more niche content.
  • Readers who visit often should get something new. If a reader visits many times a day and is caught up with the day’s most important stories, we have a unique opportunity to serve them even more of the journalism that we offer.

A major story always trumps personal interest

The routines we have developed for setting editorial meta-data are crucial for making the algorithm a tool that supports our journalistic mission. Since these signals are given significant weight in the algorithm running on the front page, a news story with high news value will rank high on the front page, even if the reader has no interest in the topic.

This helps ensure that those who primarily read about football also get news of a Ukraine invasion at the top of the front page, and a politics nerd will also be exposed to a sports story if Norway beats Brazil in a football match.

This type of dynamic ranking ensures we fulfill our societal mission while becoming more relevant to each individual.

Capturing journalism’s nuances

At the same time, we know this is not simply a question of users either being interested in sports or not. We have to avoid ending up in a situation where our internal categorisation prevents readers from discovering content they are interested in.

One of the most promising methods for addressing this is to recommend articles based on “what readers like you also have read,” also known as collaborative filtering. This means we move away from binary categorisations like “interested in sports” and “not interested in sports” and instead reflect the fact that sports articles about youth sports may be more interesting to users focused on family life than the usual sports junkies.

Similarly, an article about Haaland can, for example, be about sports, football, the Premier League, celebrities, economics, and tax policy, and may therefore be interesting to readers with no interest in football. An opinion article about FIFA and Gianni Infantino may reach those interested in both international football and politics but avoid those who only read about football in lower divisions.

The algorithm does this without really knowing what the content is about but just by analysing similarities in users’ reading patterns. In this way, the algorithm leap-frogs details about a user’s location and demography while also ignoring our internal categorisation of content in confined sections.

The risks of a continued broadcast model

The risk of not personalising news products is significant.

Digital front pages that broadcast one experience to vastly different readers will necessarily mean that many users miss the most important stories. Our readers live in a world where they are overwhelmed with information. If we are going to help them discover trustworthy journalism and relevant stories, we must embrace the idea of adapting to their needs and habits.

The danger of sticking to the old broadcast model is that we become irrelevant to an entire generation of readers. If there is one risk to democracy that we should worry about, that’s the one.

About the Authors

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.