That’s classified! Inventing a new patent taxonomy[Text matching to measure patent similarity]

Innovation researchers currently make use of various patent classification schemas, which are hard to replicate. Using machine learning techniques, we construct a transparent, replicable and adaptable patent taxonomy, and a new automated methodology for classifying patents. We contrast our new schema with existing ones using a long-run historical patent dataset. We find quantitative analyses of patent characteristics are sensitive to the choice of classification; our interpretation of regression coefficients is schema dependent. We suggest much of the innovation literature should be carefully interpreted in light of our findings.