Lots of of suspicious journals flagged by AI screening software

Researchers have recognized greater than 1,000 probably problematic open-access journals utilizing a synthetic intelligence (AI) software that screened round 15,000 titles for indicators of doubtful publishing practices.

The method, described in Science Advances on 27 August¹, might be used to assist deal with the rise in what the examine authors name “questionable open-access journals” — those who cost charges to publish papers with out doing rigorous peer evaluate or high quality checks.

Predatory publishers’ newest rip-off: bootlegged and rebranded papers

Not one of the journals flagged by the software has beforehand been on any sort of watchlist, and a few titles are owned by massive, respected publishers. Collectively, the journals have printed lots of of 1000’s of analysis papers which have acquired hundreds of thousands of citations.

The examine means that “there’s a complete group of problematic journals in plain sight which might be functioning as supposedly revered journals that actually don’t deserve that qualification”, says Jennifer Byrne, a research-integrity sleuth and most cancers researcher on the College of Sydney, Australia.

The software is on the market on-line in a closed beta model, and organizations that index journals, or publishers, can use it to evaluate their portfolios, says examine co-author Daniel Acuña, a pc scientist on the College of Colorado Boulder. However, he provides, the AI generally makes errors, and isn’t designed to interchange detailed evaluations of journals and particular person publications which may lead to a title being faraway from an index. “A human professional must be a part of the vetting course of” earlier than any motion is taken, he says.

Screening journals

The AI software can analyse an unlimited quantity of knowledge from journals’ web sites and the papers they publish, and seek for crimson flags — reminiscent of quick turnaround instances for publishing articles and excessive charges of self-citation. It additionally assesses whether or not members of a journal’s editorial board are affiliated with well-known, respected analysis establishments, and checks how clear publications are about licensing and costs. A number of of the factors used to coach the software come from best-practice steering developed by the Listing of Open Entry Journals (DOAJ), an index of open-access journals run by the non-profit DOAJ Basis in Roskilde, Denmark.

Cenyu Shen, the DOAJ’s deputy head of editorial high quality, who relies in Helsinki, says that the variety of problematic journals is rising, and that their “techniques have gotten extra subtle”. “We’re observing extra cases the place questionable publishers purchase reputable journals, or the place paper mills buy journals to publish low-quality work,” she provides. (Paper mills are companies that promote faux papers and authorships.)

I’m fearful I’ve been contacted by a predatory writer — how do I discover out?

The DOAJ’s personal high quality checks on journals are completed principally manually and are initiated solely after receiving complaints. In 2024, the listing investigated 473 journals, an increase of 40% in contrast with 2021. “The time our workforce spent on these investigations additionally grew considerably by almost 30%, to 837 hours,” says Shen.

AI instruments might assist to hurry up a few of these assessments, Acuña says. He and his colleagues educated their mannequin on 12,869 journals which might be at present listed within the DOAJ as reputable, in addition to 2,536 that the listing had flagged as violating its high quality requirements.

When the researchers requested the AI to judge 15,191 open-access journals listed within the public database Unpaywall, it recognized 1,437 journals as questionable. The workforce estimated that some 345 of those have been mistakenly flagged: they included discontinued titles, ebook sequence and journals from small, learned-society publishers. The researchers additionally discovered that the software had did not flag an extra 1,782 questionable journals, based mostly on estimates of error charges.