Close to 10% of cancer research articles have “characteristics” of paper mill publications, according to analysis by stats-sleuth Adrian Barnett (QUT) and colleagues.
That’s 260,000 papers, including 170,000 from authors at Chinese institutions – with the rate of flagged papers rising massively since the arrival of easily-accessible Large Language Models.
The authors trained a machine-learning model to distinguish real research from those it flagged as fakes, (seemingly pre-formulated sentences is a giveaway) across 1999-2024. They set-out how they screened 2.6 million papers HERE.
Findings include:
- Flagged papers were 1% at the turn of the century, rising to 15% of annual cancer research in the ‘20’s
- The top two nations for false flagged research are China (35%) and Iran (20%). The US is a distant second on the list by volume (10,500) but low by percentage (2%)
- Around 8% of papers published by giant for-profit publishers, Elsevier, Springer Nature and John Wiley were flagged, but given their volume ,this amounts to enormous numbers: Elsevier: 39,000, Springer Nature: 39,000, John Wiley: 28,000.
- The top three cancers for flagged papers were gastric (22%) and bone (21%)
- Cancer biology and fundamental research are the most common research areas, with survivorship and end-of-life among the lowest.
And yes, some 764 Australian publications were found.
But there is good news. “We hypothesise paper mills use manuscript templates with recurring features that frequently can be automatically detected,” the authors stated.