Webshops, streaming services like Spotify, and social media platforms all use recommendation systems. Some are personalized, offering suggestions based on your click behavior or previous purchases. Other systems are content-based, providing articles that align with what you've just read. NEMO Kennislink — the journalistic platform of NEMO Science Museum — will experiment in the coming months with such a content-based system that utilizes artificial intelligence (AI). It was developed by researchers from the AI, Media & Democracy Lab, a collaboration between CWI, the Amsterdam University of Applied Sciences (AUAS), and the University of Amsterdam (UvA).
Retaining readers longer
"Kennislink wants visitors to stay on the site longer and read more stories," explains Manel Slokom, postdoc at CWI's Human-Centered Data Analytics group. "Currently, many people enter through Google to read one article and then immediately leave. By providing readers with better suggestions, they hope to retain them longer. They also aim to make the website more appealing to a broader, more diverse audience — a reflection of society."
The collaborating researchers selected the algorithm for the new recommendation system, which was then trained by colleagues at AUAS. For training, the researchers used thousands of articles from NEMO Kennislink, generating 'embeddings' — mathematical representations of the content. "You want the suggestions to resemble the article someone just read, but not so much that it becomes predictable," Slokom explains. "If you keep getting stories that are almost the same, a reader will lose interest. Therefore, the system also looks at similarities in other areas: overarching theme, author, length, style."
Diverse and fair
An important aspect of designing such a system is ensuring it remains diverse and fair, Slokom emphasizes. "Kennislink wants to reach everyone in the Netherlands, regardless of background or preference. Fairness means, for example, that articles by lesser-known authors who are not widely read are also made visible by the recommendation system — not just the popular authors."
Participate as a tester
Whether the system will actually introduce more readers to the platform's extensive archive will be tested starting in April. NEMO Kennislink is seeking participants who are willing to read articles via a special test app for two months. With each story, they will receive five new recommendations. "Will they click on those? Will they return to the site the following week? We will monitor that."
How do I sign up?
Sign up for the experiment via this link (please note that the articles are in Dutch): www.nemokennislink.nl/samen-maken-we-nemo-kennislink-nog-beter
About the AI, Media & Democracy Lab
In the AI, Media & Democracy Lab, researchers collaborate on the question: how can we use artificial intelligence to make the media stronger and fairer? The goal is to deploy AI in a way that contributes to a healthy democracy.
