A revolutionary machine learning system has flagged over 250,000 cancer research papers linked to suspicious “paper mills.”
This groundbreaking research, published in BMJ, analyzed 2.6 million cancer research papers released between 1999 and 2024. Led by QUT researcher Professor Adrian Barnett from the School of Public Health and Social Work and the Australian Center for Health Services Innovation (AusHSI), the study involved international collaborators.
Researchers discovered that more than 250,000 papers exhibited descriptive patterns akin to those previously identified in studies retracted for data fabrication.
“Paper mills are operations that produce low-quality or fake scientific research. They churn out ‘research’ on an industrial scale, indicating that the extent of fraudulent activity in cancer research is far more significant than most people appreciate,” stated Professor Barnett.
Understanding how paper mills generate fraudulent research
Paper mills offer authorship and, in some cases, fully prepared, off-the-shelf scientific papers. These documents may feature recycled text, awkward language, and fabricated data and images.
“It’s possible they utilize fixed templates that can be detected by large-scale language models analyzing textual patterns,” adds Professor Barnett.
To uncover these patterns, Barnett and his team trained a language model named BERT. This advanced system is designed to detect subtle textual “fingerprints” that consistently appear in works from known paper mills.
In evaluations against validated examples, the model successfully identified suspicious documents with a 91% accuracy rate.
“Essentially, we’ve developed a scientific spam filter,” Professor Barnett explained. “Just like email systems that filter out unwanted messages, our tool flags papers resembling the style and structure of previously retracted fraudulent research.”
Rise in flagged cancer papers
A comprehensive analysis highlighted notable trends.
The number of flagged papers has surged over the past two decades, climbing from approximately 1% in the early 2000s to more than 16% in 2022.
This problematic trend spans thousands of journals published by prominent companies, including many reputable and influential scientific publications.
Research indicates that dubious papers are especially common in fields like molecular cancer biology and early-stage laboratory investigations.
Certain cancer types, including stomach, liver, bone, and lung cancers, showed particularly high rates of flagged studies.
Journals pilot AI tools
Three scientific journals are currently testing this system as a part of their editorial review process, aiming to assist editors in identifying potentially fabricated manuscripts before they’re sent for peer review.
The researchers also plan to extend the use of this tool to other scientific disciplines, anticipating that its accuracy will improve as more confirmed instances of paper mill activities are reported.
However, the team cautioned that papers flagged by the system should not be immediately deemed fraudulent. These results serve as warning signals, rather than confirmed instances of wrongdoing, necessitating further review by human experts.
The impact of fake research on patient care
“Cancer research plays a critical role in clinical trials, drug development, and patient care,” Professor Barnett emphasized.
“When fabricated studies infiltrate the evidence base, they can mislead genuine researchers and ultimately hinder progress for patients. It is crucial that we proactively address this issue.”
Source: www.sciencedaily.com