Introducing new tech into an established publisher can be tough, so Taylor & Francis have teamed up with a smart AI-focused Danish startup.
Leading academic publisher Taylor & Francis is developing natural language processing technology to help machines understand its books and journals, with the aim to enrich customers’ online experiences and create new tools to make the company more efficient.
The first step extracts topics and concepts from text in any scholarly subject domain, and shows recommendations of additional content to online users based on what they are already reading, allowing them to discover new research more easily. Further steps will lead to semantic content enrichment for more improvements in areas such as relatedness, better searches, and finding peer-reviewers and specialists on particular subjects.
This is a partnership between T&F's technology team and UNSILO, a Danish startup specialising in artificial intelligence.
What's the gap in the market?
"Technology to simplify publishing has picked up a lot in the last few years, causing significant disruption," says Nicolas Jessus, a software architect from the T&F tech team. "We started to look at how various publishing players and competitors have been approaching the problem, and noticed that a lot of the cases were being solved in a brute force way - relatedness from straight search engines, for example, or classifying content in large boxes. We figured the best way to improve on that was to harness natural language processing and machine learning, and teach the machines semantic models so that it would develop an understanding (as far as machines can currently be said to understand) of our information."
Success so far?
The team has begun by focusing on end users, trying to make relevant book suggestions. Jessus reports that "the first numbers are pretty amazing, with a conversion rate just under 20%. We're now turning our intention inwards and trying to radically increase the efficiency of our internal tools."
As always, helping a large and established publisher to introduce new tech, and to become more agile in testing it, has thrown up some challenges. "We’ve been adapting the way we are structured and working to speed up innovation," Jessus says. "We are also working with some legacy software that can be difficult to adapt - some of which are tailored to industry standard workflows that have become less relevant in a short time."
Ultimately, Jessus and his team want every piece of content ever produced by T&F to be semantically linked through machine learning to a web of knowledge - "joining everything and everyone within the academic world".
Advice to other publishers trying to innovate?
"It sounds terribly cliché, but setting up for success, really. There is nothing special about innovation - it requires a lot of managing how resources, people, and ideas are put together and unblocked, like everything else. If you want heroics and aren't prepared for struggling, you're just paying lip service to it."