A Geometry of Innovation

Abstract

We use methods from Natural Language Processing to characterize the innovative content of patents. We develop several metrics that compare inventions to existing and future innovations. The intuition guiding us is that patents dissimilar to past inventions and similar to future ones may have anticipated or started shifts in innovation topics. We find evidence that such patents have higher citations and the firms owning them grow faster and are more profitable relative to other firms. Analysis of trends suggests that innovative ideas may have gotten harder to find over time in high-innovation fields.

Keywords: NLP, Patents, Innovation, Growth

JEL Classification: O31