🙏 Huge thanks to Vashisht Bhatt for excellent research assistance, Junxi Liu for contributions to an earlier version, and the Gillmore Centre for Financial Technology at the University of Warwick for early financial support.

Couldn't have built TechText without this amazing team! 🚀

✨ Paper: "Beyond Patent Ownership: Learning About Technological Usefulness" on SSRN:

tinyurl.com/TechTextPaper

Questions about the dataset? Reach out to us!

Researchers: what would you explore with firm-tech usefulness data?

#Innovation #EconTwitter #TechPolicy

🚀 techtext.org launching soon with:
• Interactive visualizations
• Bulk data downloads
• Regular updates

This opens new research on adoption, diffusion, spillovers of innovation, and the technological positioning of the >80% of firms researchers typically can't study.

Two-stage "spy" ML approach:
🎯 Measure textual similarity between patent descriptions & firm business descriptions (TF-IDF + Sentence-BERT)
🎯 Apply positive & unlabeled ML treating patent ownership as positive signal WITHOUT assuming non-owned patents are useless

Our new dataset covers:

— 8,000+ US public firms annually on average 🏆
— 1996-2025 (nearly 3 decades)
— 50,000+ patents per year (soon 100% of USPTO grants)
— Thousands of technology categories via CPC classification
— Continuous usefulness probabilities for each firm-patent pair

Most patent-based innovation research ignores >80% 😮 of public companies because they don't patent. But non-patenting firms still use technology, benefit from spillovers, and have distinct tech profiles.
TechText solves this by going beyond patent ownership to estimate technological usefulness.