A new machine learning system has flagged more than 250,000 cancer research papers that may be connected to so-called paper mills, according to ScienceDaily.
The study, published in The BMJ, reviewed 2.6 million cancer research papers released from 1999 through 2024. It was led by QUT researcher Professor Adrian Barnett and an international team of collaborators.
The researchers said the flagged papers shared writing patterns with studies that had already been retracted because of suspected fabrication. Barnett described paper mills as businesses that sell fake or poor-quality scientific studies, sometimes offering authorship slots or even complete papers.
To identify those patterns, the team trained a BERT language model to detect subtle textual fingerprints linked to known paper mill products. In tests using verified examples, the system correctly identified suspicious papers 91% of the time.
The researchers said the number of flagged papers rose over time, reaching more than 16% in 2022. They also noted that the issue appears across thousands of journals and is especially common in areas such as molecular cancer biology, early-stage lab research, and some cancers including gastric, liver, bone and lung.
