Detection tool fights flood of fake academic papers

Academic publishers have released an online tool to spot products from “paper mills” or unauthorised outfits paid to fabricate scientific submissions, as the industry tries to boost trust by clamping down on a surge in fake research studies amplified by the use of artificial intelligence.

STM, the global body representing journal publishers, developed the tool in response to what its chief executive Caroline Sutton called “the increasing and alarming volume of materials entering scholarly communications that violate accepted research integrity.”

Paper mills cater to growing demand from researchers for published work to boost their careers. Some junior scientists will pay thousands of dollars for papers showing studies that they do not have the time or ability to carry out themselves.

An investigation by STM and its companion body, the Committee on Publication Ethics, showed that between 2 and 46 per cent of papers submitted to journals were fake. Unauthorised outfits have especially targeted biomedical research, the report showed, but the trend is evident across all scholarly areas, with many publishers reporting a rise in fraudulent submissions from China.

“The academic world is traditionally built on trust and in the past few years we have seen that process challenged by systemic fraud,” said Joris van Rossum, who has led development of the STM Integrity Hub. “The paper mill industry is a growing problem, made bigger recently by its use of more sophisticated technologies and AI-generated content.”

Publishers registered with the hub can upload a manuscript to a cloud-based tool which scans it for indications of potential fraud such as manipulated images and similarities with papers previously linked to paper mills. The version released this week is what the computer industry calls a “minimum viable product” for fraud detection and will be extended to reflect user responses.

STM is also developing a separate system that can scan different publishers for multiple submissions from paper mills. When duplicate submissions are spotted, the tool sends a notification to the journals’ integrity managers and editors so they can investigate, van Rossum added. The tool will be piloted this year.

“Journals don’t normally inform each other about what articles are submitted,” said van Rossum. “Paper mills often create a paper and offer it to multiple journals, hoping that one will accept it while the others neglect it.” 

The Integrity Hub initiative reflects an unusual level of collaboration between publishers, which are also working on their own systems. The Swiss-based Frontiers group has developed an Artificial Intelligence Review Assistant that scans submissions for signs of fraud.

But paper mills are also beginning to use AI to generate papers in which signs of fabrication are harder to spot, says Frederick Fenter, chief executive editor at Frontiers.

“We are all confronted with a growing number of fraudulent submissions,” he says. “More communication is going to be required not just among publishers but between publishers and funders and research institutions.”

A handful of independent scientific sleuths have been detecting fraudulent papers for several years. One is Elizabeth Bik, a Dutch microbiologist who has helped to develop STM’s detection tool. “It is great that publishers are now aware of this problem and that these papers are not only being retracted, but also actively screened and rejected during the manuscript submission phase,” she said.

“Preventing these papers from being published in the first place is going to send a clear message to the paper mills.” Bik added. But she warned that “AI will likely be used to generate realistic text, data sets and photographs, and it will become much harder to detect those papers”.