Anyone who has read or seen Harry Potter is familiar with the Daily Prophet. This is a newspaper that constantly updates itself, including not only pictures but also “moving images,” and adapts the content according to the specific reader.

Does this phenomenon sound familiar?

For a long time, newspapers and magazines looked the same to every visitor online. We were used to Netflix and YouTube suggesting content to us based on our viewing habits or other parameters, but print media seemed to be behind when it came to personalisation.

Major media players worldwide personalise recommendations for Web sites, apps, e-mails, and push notifications with the help of AI and recommender systems. We have written about the scientific basis behind that technology, including the work of researcher Len Feremans done at the University of Antwerp (UA), in the past.

In the first phase of research, Feremans’ study researched new techniques to improve the recommendations of regional news. He calculated the probability that someone would read an article offline, then tested that prediction against that person’s online reading behaviour.

The research concluded that combining implicit regional reading behaviour (with a number of specific selection criteria) with explicit region interests would provide the best results. On top of that, the algorithm Froomle developed to implement Len’s research in real life was made in such a way that it can cover regions as well as interests, hobbies, etc., in a very flexible way to cover as many use cases as possible.

Mediahuis is one of the leading media companies that uses personalisation technology across its different newspapers. Mediahuis newspapers have an extensive regional content offering, but until recently did not personalise this offering. All users received the same recent regional news as recommendations.

By applying Froomle technology in practice, we learned many interesting insights. We started matching recent regional news with the user’s reading history. In this step, we started personalising the regional recommendations for 75% of all users, coming from 0%.

Froomle started suggesting regional news based on extensive user profiles in the second phase, based on the UA Research. User profiles for all recurring users are calculated based on the regional articles they read or marked as an interest. These user profiles are used on Web site recommendations and in outbound push and e-mail recommendations. By doing this, it’s easier to recommend items based on the long-term interests of a user while combining with short-term intent.

While offering an out-of-the-box solution for personalising online channels or push notifications, the functionality is developed to allow for flexible customisation by the newsroom. The end goal is to help newsrooms optimise their content distribution processes, while giving the team control and options for tailoring for specific situations.

For one newspaper, a “region” is a province; another medium might consider three streets a region. Newspapers such as Het Nieuwsblad publish dozens of articles about Ghent every day but perhaps one about a small village in Limburg every week. Our current implementation allows us to work with these different use cases.

Currently, the implementation is based on a suggestion by Froomle, but the customer can tweak a lot of parameters, too. Based on the newsroom preferences, editors can tailor the audience selection to each situation. There are numerous options: For example, editors can choose the maximum number of pushes a user can receive per day or hour, decide the minimum interest in a specific topic before determining eligibility for a push notification, use explicit interests in the selection of the audience, or include neighbouring regions or related interests.

In other words, automation, flexibility, and personalisation aren’t just for Big Tech giants anymore. Every news publishing media can define the optimal balance between hyper-targeting (high click-through rates) and broad audiences (higher absolute clicks). Thanks to the algorithms suggested by academic research, they can automatically increase the click-through rate by finding that sweet spot while giving editors more autonomy.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .