Patent-to-Patent Similarity: A Vector Space Model

Abstract

Current measures of patent similarity rely on the manual classification of patents into taxonomies. In this project, we leverage information retrieval theory and Big Data methods to develop a machine-automated measure of patent-to-patent similarity. We validate the measure and demonstrate that it significantly improves upon existing patent classification systems. Moreover, we illustrate how a pairwise similarity comparison of any and every two patents in the USPTO patent space can open new avenues of research in economics, management, and public policy. We make the data available for future scholarship through the Patent Research Foundation.