Reliance on science by inventors: Hybrid extraction of in?text patent?to?article citations

We curate and characterize a complete set of citations from patents to scientific articles, including 16.8 million the full text USPTO EPO patents. Combining hand-tuned heuristics GROBID machine-learning package, we achieve much higher performance than machine learning alone. Recall is evaluated with 5939 randomly sampled, cross-verified “known good” citations, which authors have never seen. At 99.4% precision, recall rates 78% for test 88% references specified without mistakes. compare these “in-text” those on front page In-text are more diverse temporally, geographically, topically; moreover, they less self-referential likely be copied one patent next. dropped two-thirds all patent-to-article half century ago about one-third today. In replicating two articles that use only front-page show failing capture in-text leads understating role academic science in commercial invention. All known-good set, source code available at http://relianceonscience.org.