New Ways to Determine Patent Novelty

Jonathan Ashtor (now at Paul, Weiss) has completed a few quality empirical studies in the past. His new foray is a new and creative way to determine novelty. It's on SSRN, and the abstract is here:

I construct a measure of patent novelty based on linguistic analysis of claim text. Specifically, I employ advanced computational linguistic techniques to analyze the claims of all U.S. patents issued from 1976-2014, nearly 5 million patents in total. I use the resulting model to measure the similarity of each patented invention to all others in its technology-temporal cohort. Then, I validate the resulting measure using multiple established proxies for novelty, as well as actual USPTO Office Action rejections on grounds of lack of novelty or obviousness. I also analyze a set of pioneering patents and find that they have substantially and significantly higher novelty measures than other patents.

Using this measure, I study the relationship of novelty to patent value and cumulative innovation. I find significant correlations between novelty and patent value, as measured by returns to firm innovation and stock market responses to patent issuance. I also find strong correlations between novelty and cumulative innovation, as measured by forward citations. Furthermore, I find that patents of greater novelty give rise to more important citations, as they are more frequently cited in Office Action rejections of future patents for lack of novelty or obviousness. I also investigate how novelty relates to the USPTO examination process. In particular, I find that novelty is an inherent feature of a patented invention, which can be measured based on the claim text of either an issued patent or an early-stage patent application.

Next, I use this measure to analyze the characteristics of novel patents. I find that novelty is an effective dimension along which to stratify patents, as key patent characteristics vary significantly across the distribution of novelty measures. Moreover, novel patents are closely linked to basic scientific research, as measured by public grant funding and citations to non-patent scientific literature.

Finally, I use this measure to observe trends in novelty over a forty-year timespan of American innovation. This reveals a noticeable, albeit slight, trend in novelty in certain technology fields in recent years, which corresponds to technological maturation in those sectors.

 I'm skeptical of measures of patent quality by claim language only, but I like how he has used office actions to validate the measure. I think people will have to study this to see how it holds up, but I think it's an interesting and creative first step toward objectively judging quality.