An index measures the extent to which a scientific article "speaks like a patent" to identify valuable research early

The Translation Readiness Index scores the lexical proximity between an article abstract and publications already associated with patents. The model achieves 78% correct rankings.

Identifying commercially promising research before patenting is a big part of the work of investors and university valuation departments. A machine learning tool called the Translation Readiness Index (TRI) proposes to speed up this sorting by assigning each article a resemblance score to the language of patents — months or even years before an agreement, filing or business creation reveals its potential.

The IRR only analyzes the title and abstract: it measures the similarity of the vocabulary used to that of publications already matched to patents. Developed by researchers at the Sydney-based data analysis firm League of Scholars, it was trained on 20,610 papers, 9,431 of which were associated with a patent, divided into five classifiers. The best of them has a 78% chance of placing a patent-related article ahead of a comparable article that is not. Texts subsequently cited in patents more often use terms such as "prototype", "device" or "design".

The limits are assumed: by sticking to the title and the summary, the tool does not judge either the data or the underlying results. The work has also been pre-printed on arXiv and has not yet been peer-reviewed. Finally, what is patentable is not necessarily commercially viable, which immediately limits the scope of such a measure for an investor.