‘Unseen’ efforts to catch paper mill outputs bear fruit

Unseen efforts to combat academic fraud are intercepting about 1,000 submissions from paper mills every month as publishers move towards more “upstream” detection of research integrity breaches, says a new report.

While the retraction of scientific papers can often generate significant media attention, publishers are increasingly focused  on “catching problems before publication rather than correcting the published record”, explains a study commissioned by the International Association of Scientific, Technical and Medical Publishers (STM).

That “strategic shift” follows a change in the “nature and scale of potential breaches” seen by publishers who previously encountered “sporadic intentional manipulation” and “individual lapses in judgement” but now face “large-scale operations selling manufactured manuscripts, AI systems capable of generating plausible but fabricated research, and coordinated networks that span journals and borders”, explains the report by STM and Research Consulting, published on 13 January.

That has been combined with growing publication rates, with 5.7 million articles, reviews and conference papers appearing in 2024, up from 3.9 million five years earlier, according to data sourced by the Digital Science system Dimensions.

Detailing the growth of research integrity teams within publishers – which exceed 100 people at some larger companies – the report also outlines some of the collaborative initiatives used to improve screening and widen access to detection tools and expertise, including the STM Integrity Hub, which was launched in May 2022.

That collaboration integrates 15 independent tools for finding suspect papers and is used by more than 35 publishers to screen about 125,000 papers every month, detecting an estimated 1,000 suspect papers a month.

Speaking to Times Higher Education, Chris Graf, research integrity director at Springer Nature, said he hoped the study, which drew on insights from 18 research integrity experts across 13 organisations, would highlight how publishers have scaled up investments to stop suspect papers from entering the scientific literature, given this work is largely unseen.

“There is an awful lot going on but it’s not very visible, even to those working in the publishing world,” said Graf, who said his research integrity team had tripled in size in recent years.

“I have 55 people in my team but we’ve also added people in teams adjacent to integrity: technology, legal, product management and communications,” he said, adding: “This work has always been part of what we’ve done but the threats have changed.”

Industry-wide efforts to combat research fraud represented a new level of collaboration between publishers that he had not anticipated, continued Graf. “It is quite remarkable – I have never been part of a collaboration which is as universal and understood to be critical [to the sector’s future] as the STM Integrity Hub,” he explained.

The hub shares technology developed by larger publishers such as Elsevier and Springer Nature but also provided access to detection tools created by smaller third party providers, Graf said.

“Some of this technology was quite start-uppy at first but now it’s suitable for use at scale,” he said. “It’s so critical that we keep on collaborating and sharing information because you can bet your bottom dollar that paper mills will continue to evolve so we need to be able to spot those [fraud] signals so we can respond,” he said.

Despite the STM Integrity Hub’s success, however, the report notes its participation involved a “limited proportion of the sector’s thousands of publishers globally”.

“Extending collaborative benefits to smaller publishers and those operating outside major publishing centres represents a significant scaling opportunity,” it recommends.

While the report warns “increasingly sophisticated capabilities” are required to keep pace with “increasingly sophisticated threats”, Graf said he was hopeful that publishers could stay ahead of AI-assisted research fraud.

“There are different ways to assure trust – there is metadata, trust markers where you have a research fingerprint that can be identified. And increasingly we are going beyond working with publishers,” he said, stating publishers are even working with those who manufacture the machines that produce Western blots, used in protein detection, to help improve verification of results.

“This is publishers getting really collaborative and that can be really powerful,” he said.

jack.grove@timeshighereducation.com