About 1,000 of a set of 15,000 free access scientific journals seem to exist mainly to extract the costs of naive academics.
A trio of computer scientists from the University of Colorado Boulder, the University of Syracuse and the Chinese Technology Institute (EIT) has reached this figure after having built a classification of automatic learning to help identify “doubtful” journals, then carry out a human examination of the results – because AI is short of itself.
A questionable newspaper is the one that violates best practices and has low editorial standards, which are mainly existing to have academics to pay high costs so that their work appears in a publication that does not provide the expected editorial exam.
As detailed in a research document published in Science Advances, “estimate of the predictability of questionable free access journals”, scientific journals prior to the 1990s had tended to be closed, available only through subscriptions paid by institutions.
The free access movement has changed this dynamic. It dates back to the 1990s, because the free software movement was gaining momentum, when researchers sought to extend the availability of academic research. A consequence of this transition, however, is that the costs associated with the revision of peers and publication have been displaced by organizations subscribed to the authors.
“The free access movement has been planned to correct this lack of accessibility by modifying the payment model,” explains the document. “Places with free access ask the authors to pay directly rather than ask universities or libraries to subscribe, allowing scientists to keep their copyright.”
Free access scientific edition is now widely accepted. For example, a memorandum in 2022 of the Board of the White House and Technological Policy Office ordered American agencies to offer a plan by the end of 2025 to make research supported by taxpayers.
But change towards free access has led to the proliferation of questionable scientific publications. For more than a decade, researchers have raised concerns about predatory and diverted journals (PDF).
The authors attribute Jeffrey Beall, librarian at the University of Colorado, to have applied the term “predatory edition” in 2009 to suspect journals that try to extract the costs of the authors without editorial review services. An archived version of the Beall’s potentially predatory journals and publishers can still be found. The problem with an approach based on the list is that scam journals can easily modify their names and websites.
In light of these problems, Daniel Acuña (UC Boulder), Han Zhuang (EIT) and Lizheng Liang (Syracuse), have decided to see if an AI model could be able to help separate legitimate publications from those questionable using detectable characteristics (Eg Authors who frequently cite their own work).
“Science is progressing based on the work of others,” said Acuña The register in an email. “Bad science pollutes the scientific landscape with unusable results. Doubtful journals publish almost anything and, therefore, the science they have is not reliable.
“What I hope to accomplish is helping to get rid of this bad science by helping in a proactive way to mark suspected journals so that professionals (who are rare) can focus their efforts on what is most important.”
Acuña is also the founder of Reviewerzero AI, a service that uses AI to detect research integrity problems.
Generating a set of data of nearly 200,000 free access journals, the three computer scientists settled on a set of 15,191 of them.
They formed a classifier model to identify questionable journals and when they executed it on the shooting of 15 191, the model reported 1,437 titles. But the model missed the brand about a quarter of the time, on the basis of a subsequent human review.
“About 1,092 should be truly questionable, ~ 345 are false positives (24% of the assembly reported) and ~ 1,782 problematic journals would remain unteashed (false negative),” said the newspaper.
“At a wider level, our technique can be adapted,” said Acuña. “If we care about a lot of false positives, we can report more rigorously.” He underlined a passage in the document which says in a stricter framework, only five false alarms out of 240 would be expected.
Acuña has added that if many AI applications are today aimed at complete automation, “for questions as delicate as the one we examine here, AI is not yet there, but it helps a lot.”
The authors are not yet ready to appoint and be ashamed of questionable journals – this could invite a legal challenge.
“We hope to collaborate with indexing services and help deemed publishers who could be concerned about the deterioration of their journals,” said Acuña. “We could make it available in the near future for scientists before submitting to a review.” ®