Credit: Image by Yulia Grigoryaya on Shutterstock

A year after announcing it had received funding in the first round of Google's Digital News Initiative, Swiss daily newspaper Neue Zürcher Zeitung (NZZ) is getting ready to test the beta version of the product it has been developing over the last six months: a personalised news engine named the Companion app.

The app uses a personalisation algorithm based on machine learning to provide readers with a stream of relevant NZZ stories and understand their consumption habits. Its main goal is to help people "find the right news we've produced during the day with less effort", explained Rouven Leuener, NZZ's head of product development.

It needs to have some kind of mixture because we do not want you to feel like you are in a filter bubbleRouven Leuener, NZZ

But as opposed to taking the usual route of recommending a story to someone because it is similar to an article they have already read, the app's approach to personalisation will be based on the meaning of an article and it will "always include an element of surprise", he added.

This means that even if, based on their consumption history, someone has not expressed an interest in sports, their stream will include news about big, important stories related to sports, for example a scandal or an athlete winning an Olympic medal. "It needs to have some kind of mixture because we do not want you to feel like you are in a filter bubble."

The algorithm currently personalises the stream for each individual by awarding each story an editorial score – Leuener said it was too early to disclose the variables the score takes into account at this point – and linking it to the user's profile. At the moment, the stream has 25 stories and refreshes every hour, when two to three articles are removed from it and replaced by new ones.

Since the 200 or so people who will be testing the app over the next four months are a mix of current NZZ subscribers and new readers, those who already have an account will have an existing history that shows roughly what type of stories they are interested in and when they tend to consume content.

For unregistered users, who will be given NZZ subscriptions for the duration of the testing period, the algorithm will be able to learn about their interests and behaviour in a couple of days after they have started using the app and present them with a relevant stream of articles. "This is where we are investing a lot of time," Leuener explained, referring to the algorithm's ability to learn quickly.

Those testing Companion will have to provide regular feedback on the look and feel of the app and the quality of their personalised news streams, such as what type of stories they would like to see more or less of, or whether their stream was better one day compared to another. This will help the team, which includes Leuener, NZZ chief product officer Anita Zielina and four other staff members, further develop the product and add new features before the final version of the app is scheduled to be released at the end of July.

"We are considering many features for the algorithm, such as time spent or scrolling, we could measure lots, we just don't know yet the impact it should have on your personalised stream.

"This is something we will develop over the last months and if something doesn't make the algorithm better or cleverer, we'll throw it out, but we won't know which features we will incorporate until we have the feedback from the testers."

The app could also be able to recognise if a reader prefers to consume a certain type of stories at a particular time of the day, such as financial updates in the morning, and adjust their stream accordingly.

Personalisation is a "very crucial topic" and it shouldn't be confused with editorial curation, which should remain "untouched", Leuener added. The Companion app's personalised feed is separate to the stream of stories curated by editors, which are featured in a separate tab. There has been a notable change in how personalisation is perceived in the last few months, as people have come to understand the effect of filter bubbles, particularly on Facebook.

"I think we need to take into account and be aware where personalisation is taking place and where it isn't," Leuener said.

"There's a big difference between personalisation on Facebook and ours. We don't necessarily want you to read more articles or to only provide you with the information you're more likely to click on.

"Our approach is different. We only have a few hundred articles a day, so it's a more limited amount compared to Facebook's enormous content base, and the only service we provide is that you will be able to find the relevant stories with less effort."

It has not yet been determined if the Companion app will be released as a standalone product or if the algorithm will come to influence the outlet's overall approach to personalisation in its other digital products, such as the NZZ news app or the website.

For us, personalisation is not only focused on content recommendations, but can have significant impact on how we guide users through our paid offeringsAnita Zielina, NZZ

"The reason why we designed our DNI project more like a data layer than like a standalone app is because we can imagine it to become part of various NZZ products," Zielina told Journalism.co.uk in an email. "If our tests show that our customers like personalisation to be part of their user experience, we will think about new applications."

Other ways the algorithm could be incorporated into NZZ's editorial offering could be by surfacing relevant archival content for readers or personalising push alerts in an app or on the website to reach a specific audience, she added.

"As a subscription oriented media business, we need to create products and services that people are willing to pay for. For us, personalisation is not only focused on content recommendations, but can have significant impact on how we guide users through our paid offerings.

"Media organisations can learn quite a lot from e-commerce businesses and retailers that have been experimenting with increasing conversion rates through personalisation of user journeys, landing pages and subscription prompts," Zielina said.

Free daily newsletter

If you like our news and feature articles, you can sign up to receive our free daily (Mon-Fri) email newsletter (mobile friendly).