A publication of the Centre for Advancing Journalism, University of Melbourne

Culture

Perfect match: Recommending children’s books in the age of algorithms

Can an algorithm make good book recommendations? In this third article in a series on e-books, Petra Stock reports on the way e-book platforms suggest what to read next.

Perfect match: Recommending children’s books in the age of algorithms

Tye Cattanach works at specialty children's shop Readings Kids Photo: Petra Stock

Words by Petra Stock
 

Helping children find a book they’ll enjoy reading is the most fun part of Tye Cattanach’s job at specialty children’s bookshop Readings Kids.

“If a kid says to me, ‘I’m playing Assassin’s Creed and I love it’, then you sort of get a feel for the fact that historical fiction is going to work for them,” she said.

Recommending books is a specialised skill for Ms Cattanach, a qualified teacher-librarian and with a background in children’s publishing.

She will often start out asking the child whether they actually like to read. Many of them don’t, she said, or are sort of undecided about it.

To find the right book, she’ll ask lots of questions to find out what the child is interested in. Do they have any hobbies? What video games do they play?

It’s a skill that requires broad awareness of children’s popular culture as well as deep knowledge of the books available.

The most fun part of Tye Cattanach’s job is helping children find a book they’ll enjoy reading<br>Photo: Petra Stock

The most fun part of Tye Cattanach’s job is helping children find a book they’ll enjoy reading
Photo: Petra Stock

Ms Cattanach helps adults too.

During lockdowns, the phones at Readings Kids were relentless, she said. Often it was interstate grandparents calling, wanting help to find something to read with their Melbourne-based grandchild over Zoom or FaceTime.

Can an algorithm make good book recommendations?

Meanwhile, on the e-book streaming app Epic, the job of recommending new books to children is done by a computer algorithm.

The app collects data on reading behaviour which it then analyses to offer new suggestions for what to read from its digital library of 40,000 books. The automated suggestions work in a similar way to music or television show recommendations made on Spotify or Netflix.

But how well can an algorithm recommend the right book for a child?

Dr Luci Pangrazio is a research fellow in digital literacies at Deakin University. She investigates how young people’s data is being collected and used, and the implications for their privacy and freedom.

Dr Pangrazio explained these types of recommendation algorithms often draw on past information – like what children have read previously – to determine what books would be a good fit.

“The way these algorithms work is they use that historical data… not only on how old the child is, where they live, but also what they’ve read before… [The app] would have very detailed information on how long they read that book for, whether they got to the end,” she said.

But this approach may not necessarily be a reliable way to choose new books for a child to read, she said.

For instance, the algorithm doesn’t know if a child read a book, but didn’t like it. It can also be hard to move out of a particular genre. “If you like fantasy novels, you’re going to be recommended more fantasy novels,” Dr Pangrazio said.

Dr Pangrazio said that other platforms like YouTube had begun to address some of these limitations.

“Interestingly they actually counter that kind of bubble by throwing in diverse choices into their algorithms – actually built into their designs – to make sure you get something out of left field because they know that people, their viewers, like that diversity,” she said.

What happens without a human in the loop?

Monash University professor Neil Selywn has been researching education and technology for 25 years, and has a particular interest in the rise of platforms in schools and data analytics.

Professor Selwyn said the quality of these computer-based recommendations depends on the books available on the platform, and the ways children interact with the app.

He described Epic as “better than nothing” as a way to provide children with access to books to read during lockdowns, however he questioned the quality of many of the books in its library.

“It’s like a very low quality Netflix for books, with films you’ve never heard of,” Professor Selwyn said.

His son’s class used Epic during COVID-19 lockdowns. Professor Selwyn said it was interesting to see how children interacted with the platform, and how quickly they learned to manipulate the app’s recommendation algorithm.

For example, children will try to click through books as quickly as the platform will allow, in order to collect points, or get a new avatar. This type of interaction – where children aren’t really engaging with the books, but rather maximising their points – leads to bogus data and feeds bogus recommendations, he said.

According to Professor Selwyn, this temptation to manipulate the system reflects the way a gamer might approach reading books. Gamification elements built into the platform include rewards like points and badges for reading books or completing quizzes.

“Without a human in the loop”, such as a teacher or parent providing oversight, the risk is that children get caught in a loop of being recommended substandard content, Professor Selwyn said.

Epic tracks data on what children are reading on the platform to identify trends and what children are interested in.

“All major media companies are always looking for what’s the next big thing that kids are going to be interested in,” said Epic co-founder Kevin Donahue in an interview with yahoo!finance.

“We have data from a billion books that have been read in our service,” he said.

Epic’s privacy policy states the company collects information on how children interact with the app.

“For example, we collect information about how many books your child user has read, the length of time to complete a book, responses to quizzes and the child’s reading abilities or interests,” the policy states.

The app uses the data for generating personalised reading recommendations based on the child’s interests and reading level, among other uses. The company states it does not use the data for targeted advertising or marketing.

In 2018, Epic began using aggregated data on children’s likes and search terms for a new purpose.

Having amassed information on millions of children’s reading behaviour, likes and searches on the app, Epic began using analytics and insights from the data to inform the production of its own line of books called Epic Originals, the Wall Street Journal reported.

Characters and content in the books are informed by data on what’s trending from the app. The series includes Cat Ninja, Unicorn Island and Cosmic Pizza Party.

Epic is not the only e-book company to collect and use reader data.

Amazon collects and stores data from its Kindle users, tracking details like page turns, highlighted extracts, searched words, and when sections of books were copied and pasted, The Guardian reported.

Amazon told The Guardian it did not share individual’s data, but used the information in aggregated form “to provide customers with products and services, pay content providers and improve the reading and shopping experience”.

Neither Epic nor Amazon responded to questions about the data they collect on readers and the way this was used by their recommendation algorithms.

Humans know the book, know the reader

Dr Susan La Marca, executive officer of the School Library Association of Victoria said an underlying problem with e-book recommendation algorithms is the metadata or information that categorises each book.

“Who did the metadata? Who actually decided: this book is fantasy; this book is for a 10-year-old?”

She said in many cases the metadata was originally created for a different purpose, and may have been done by people who hadn’t actually read the book.

That’s why, she said, algorithms might throw up some good suggestions, and then a whole lot of other recommendations which don’t make any sense.

An algorithm is only as good as the data it relies on. Whereas the human skill of recommending books requires librarians to read everything, and to have conversations with and know their students, Dr La Marca said.

“You need to know the book, and you need to know the reader,” she said.

Despite the adage, judging a book by its tangible cover is often helpful in this process too.

“In a physical library, you can pull the book off the shelf, and you can look at the cover and you can flip through it and you can perhaps get a sense of it,” Dr La Marca said.

“In an e-book library… It’s really hard to choose.”

Dr La Marca said the human skill of recommending books – matching the right book to the reader – is undervalued and increasingly under threat as schools have less money for staffing libraries.

Scarce resources were stretched even further and funds  diverted to provide digital devices and access options during the pandemic, she said.

While governments may spruik money for school libraries, it is often spent on buildings and resources rather than staffing them adequately.

“So you’ll go to most schools, and they’ll have something they can show you is the library, but they won’t necessarily have a trained professional running that space,” Dr La Marca said.

Tye Cattanach says the face-to-face element is important when recommending books for younger children<br>Photo: Petra Stock

Tye Cattanach says the face-to-face element is important when recommending books for younger children
Photo: Petra Stock

Ms Cattanach said many children are familiar with the way recommendation algorithms work, and use them on other platforms such as TikTok.

But when it comes to books, she said the human element is especially important, especially for younger children.

“I see some very confident eight-year-olds who know exactly what they want to read. That’s rare,” Ms Cattanach said.

“The most common thing I see is kids who are just like, ‘I don’t know what I want to read, I don’t know what I’m interested in.”

So, what would she recommend for the child who plays Assassin’s Creed?

Potentially Ranger’s Apprentice by John Flanagan for its narrative arc, or the Sabriel series by Garth Nix, but it would really depend on the individual, Ms Cattanach said.

“You have to have that conversation there and help them.”

About The Citizen

THE CITIZEN is a publication of the Centre for Advancing Journalism. It has several aims. Foremost, it is a teaching tool that showcases the work of the students in the University of Melbourne’s Master of Journalism and Master of International Journalism programs, giving them real-world experience in working for publication and to deadline. Find out more →

  • Editor: Jo Chandler
  • Reporter: Petra Stock
  • Audio & Video editor: Louisa Lim
  • Data editor: Craig Butt
  • Editor-In-Chief: Andrew Dodd
  • Business editor: Lucy Smy
Winner — BEST PUBLICATION 2016 Ossie Awards