Publishers leverage AI, personae in personalisation strategies

By Paula Felps

INMA

Nashville, Tennessee, United States

Connect      

By Michelle Palmer Jones

INMA

Nashville, Tennessee, United States

Connect      

By Elaine Sung

INMA

Memphis, Tennessee, United States

Connect      

As user needs continue to evolve, media companies are experimenting with personalisation to better understand and serve readers as individuals.

“It’s not just content that we can personalise: Think formats and sequencing of content,” INMA Product Initiative Lead Jodie Hopperton wrote in her takeaways from the Product Innovation Master Class. “It should be across all platforms to keep a seamless experience. And we should think about how a single reader has different needs at different times, including their aspiration as well as their history.”

Media company leaders and experts shared their work around personalisation and valuable insights into how they approach this complex subject during the recent event.

Athena Institute

While personalisation right now focuses on different types of readers and the kind of content that appeals to them, the research of Jaron Harambam, assistant professor of participatory AI at the Athena Institute in the Netherlands, looks at how to personalise selections for one reader at different times of the day or for different moods.

For the last five years, Harambam has been developing what he calls news recommender personae. They revolve around the idea that there isn’t such a thing as a “stable me.”

“Are you and am I the same person every day of the week, every moment of the day?” he asked. “I don’t think so.”

News recommender personae can enable user-led personalisation, Jaron Harambam, assistant professor of participatory AI at the Athena Institute, said.
News recommender personae can enable user-led personalisation, Jaron Harambam, assistant professor of participatory AI at the Athena Institute, said.

The news recommender personae are designed to cater to those wide-ranging moods and interests. In giving users a choice over which personae they would like to engage with, it also gives them a more personalised experience.

He used the example of visiting CNN’s Web site and being able to select which persona you’d like to have curate the content you consume. The “explorer” persona would offer “news that is personalised to you but opens up your own horizon,” while the “expert” persona would offer “highly specialised information that is dependent on the things you already like” and will take you deeper into those topics.

By putting a human face on recommendation algorithms, Harambam said it helped people not only intuitively understand their specific news selections, but also allows them to choose news based on their own mood and interests. His research in this area continues.

“News recommender personae actually enable and offer the option for people to pick different kinds of personalisation algorithms to give users more choice in the specific selections they get,” he said.

Ekstra Bladet

As the next big thing in digital publishing — and in virtually every other industry as well — Artificial Intelligence is having a moment. Kasper Lindskow, head of research and innovation at Denmark’s Ekstra Bladet, shared insights into the company's efforts to use AI to create a “deeper, wider, and richer news experience.”

Ekstra Bladet uses its recommender system to connect readers to news, Kasper Lindskow, head of research and innovation, said.
Ekstra Bladet uses its recommender system to connect readers to news, Kasper Lindskow, head of research and innovation, said.

Setting a goal to create a more relevant, engaging, and informative user experience, Ekstra Bladet wanted to use AI systems that were aligned with its values and were independent of technology giants. It wanted to maintain control in-house rather than delegating the tools to external vendors.

“At the same time we were hoping to contribute to a healthy norm for how to use AI in media, because lots of questions are still unanswered,” he said.

After seeking funding from external foundations and collaborating with partners, Ekstra Bladet was able to build a recommender system for news. It uses natural language processing (NLP) algorithms to read news articles and understand their content. The idea was to build an internal team but also have external partners to help develop the proper algorithms.

Ultimately, it focused on building recommender systems, adapting algorithms to suit its needs and then implementing the NLP systems. That, he said, has transformed the news experience.

“Were sort of classical news publisher that’s mainly text-driven,” he said. “So it makes sense for us to focus on recommender systems because that’s all about connecting readers to news, which is what we’ve always done. Recommender systems just allow us to do it in a more granular fashion.”

The Wall Street Journal

The Wall Street Journal is all in on personalisation. They understand user needs are constantly changing and in order to keep on top of what users need, WSJ is finding they have to be extremely nimble in how they offer personalisation to users.

Sharon Denning, senior vice president of product and customer experience, said WSJ is trying to bridge the gap between what companies are trying to do with personalisation and what’s actually landing for their users.

The Wall Street Journal is trying to bridge the gap between what companies are trying to do with personalisation and what’s actually landing for their users, Sharon Denning, senior vice president of product and customer experience, said.
The Wall Street Journal is trying to bridge the gap between what companies are trying to do with personalisation and what’s actually landing for their users, Sharon Denning, senior vice president of product and customer experience, said.

The idea of active versus passive personalisation is one WSJ is looking into. They’re testing the idea of helping users understand what’s happening behind the scenes without overwhelming them with all the details. There’s also a balance needed between creating a really relevant, personalised experience for the user and for the company to increase engagement without losing sight of the ultimate goals.

“One of the things I feel is the most important is that it actually does tie back to our core mission, which is bringing the information that matters to readers where and when they need it,” Denning said. “And so we have to kind of remember that personalisation is a tool for us in terms of doing whats most important for us and kind of our mission and why were here.”

VK Media

Josefin Svensson, data scientist at VK Media in northern Sweden, said the company’s in-house personalisation service “Matcharn” (the Matcher) builds target groups based on what users have been reading, so when they sign in, they get a personalised recommendation that VK Media calls Powerpuffs.

The “normal” puff is a general teaser, but the personalised puff has a marker heading “For You” at the top, followed by headlines that fall into that target group. Impressions and clicks are tracked off the puffs.

VK Media's personalisation service “Matcharn” (the Matcher) builds target groups based on what users have been reading, Josefin Svensson, data scientist, said.
VK Media's personalisation service “Matcharn” (the Matcher) builds target groups based on what users have been reading, Josefin Svensson, data scientist, said.

Target groups are built based on pageview data (what stories a user has read) and article data on topics to be built. A user’s “reading value” is calculated, based on what topic he has read, what kind of story, and the size of the story. Similar users are collected into a target group.

Now that Matcharn has the user’s interests, its algorithm dictates the order of the headlines, avoids duplicate stories from the site’s first page, and ensures that each time a user signs in, they see a different story.

Multiple departments participated in this first project, which was built in-house in a couple of months and evolved from there, Svensson said: “We worked in this project with continuous evaluation and it proved to be very useful because we could do some experiments and look at the results and find out if it was performing good or not and make adjustments accordingly instead of just trying to guess what would give us the best results.”

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.
x

I ACCEPT