Drivers of the decrease of patent similarities from 1976 to 2021

The citation network of patents citing prior art arises from the legal obligation of patent applicants to properly disclose their invention. One way to study the relationship between current patents and their antecedents is by analyzing the similarity between the textual elements of patents. Patent similarity indicators have been constantly decreasing since the mid-70s. The aim of this work is to investigate the drivers of this downward trend through a general additive model and contextually propose a computationally efficient way to derive the similarity scores across pairs of patent citations leveraging on state-of-the-art tools in Natural Language Processing. We found that by using this non-linear modelling technique we are able to distinguish between distinct, temporally varying drivers of the patent similarity levels that accounts for more variation in the data ($R^2\sim 18\%$) in comparison to the previous literature. Moreover, with such corrections in place, we conclude that the trend in similarity shows a different pattern than the one presented in previous studies.