AI SCIENCE
A spam filter for scientific papers
A new AI tool has flagged more than 250,000 cancer research papers with writing patterns linked to suspected “paper mills”.
Researchers analysed 2.6 million cancer papers published between 1999 and 2024. The study appeared in The BMJ.
Paper mills sell authorship spots or complete research papers, sometimes using copied text, fake data or altered images.
The team trained a language model called BERT to spot repeated writing patterns found in previously retracted papers.
In testing, it detected suspicious papers with 91% accuracy.
The number of flagged studies rose from around 1% in the early 2000s to more than 16% in 2022.
They appeared across thousands of journals, including some run by major publishers.
The key findings:
More than 250,000 papers were flagged.
Suspicious papers peaked at over 16% in 2022.
The tool supports editors but does not replace expert checks.
The copy-paste laboratory
Three journals are now testing the tool before papers are sent for peer review. Researchers also hope to use it in other areas of science.
The team stressed that a flagged paper is not automatically fake. It simply means the study needs closer human review.
Imagine working on a paper for 7 years and then it gets flagged as fake. - MV