Pictures Worth A Thousand Words? AI for Design Patents

Abstract

Patent examination process is crucial for firms and policymakers alike. From a firm's perspective, the duration of patent examination process and grant decision carries important strategic, legal, and financial implications. At the same time, policymakers are interested to optimize the allocation of scarce examination resources and improve the reliability of patent grant assignments. I propose using Image Analysis to enhance and standardize patent examination procedure. Using granted design patents for mobile phones issued by the U.S. Patent and Trademark Office (USPTO) from 1990-2018 as an exemplary case, I extract prior art contribution measures and develop an algorithm to assign new design patents. Results show that contrary to prior research on technological patents, text-based similarity and novelty measures do not perform well in the case of design patents. Next, results suggest that a high level of prior art contribution would have a shorter patent examination process. Lastly, evidence exhibits that the trained neural network has a 94 percent accuracy rate in automatically classifying new patent designs. I draw policy and managerial implications.