A machine-learning tool that predicts whether a scientific paper will eventually be cited in a patent is drawing attention from investors and technology-transfer offices, according to a report published today in Nature.
The Translation Readiness Index (TRI), developed by researchers at the data-analytics firm League of Scholars in Sydney, performs linguistic analysis of a paper's title and abstract. It measures how similar a paper's vocabulary is to publications that have previously been paired with patents — detecting terms like prototype, device, and design that correlate with commercial potential.
In testing, the best-performing TRI classifier had a 78% chance of ranking a patent-linked paper above a comparable non-patent paper. When applied to 100 highest-ranked papers from the University of Western Australia, 83 had industry-affiliated co-authors and 34 involved authors who had previously patented research.
It is a new way of triaging or ranking research, says co-author Paul McCarthy. It might help to uncover unexpected gems.
TRI joins a growing field of research-scouting tools. Cornell University uses a tool called Haystack to scan up to 13,000 papers per year for its technology-transfer team. But researchers caution that patentability does not guarantee commercial viability, and the tool analyzes only titles and abstracts — not underlying data or results.
