Modeling an indicator for statutory patent novelty

Abstract

Novelty is considered a conditio sine qua non for the grant of a patent by most relevant patent authorities and in U.S. patent law defined by 35 U S C. §102. Previous attempts to operationalize patent novelty have been mostly based on theoretical principles, such as recombination theory, and have not estimated novelty according to data that officially determines whether sufficient novelty is present in an application. To overcome this gap, this study analyzes whether established measures of patent novelty are capable of predicting the rejection of an application based on lack of novelty. Furthermore, this study applies a combination of sophisticated unsupervised and supervised machine learning techniques to patent data and provides the possibility to estimate statutory novelty practiced by the USPTO by a modeled indicator. Measuring such statutory novelty would give applicants a tremendous competitive advantage to pursue patent strategies offensively and/or defensively. For example, the indicator allows companies – particularly small and medium-sized companies with limited resources – to assess the novelty and therefore the patentability of one’s own invention. Large companies, most of which are more likely to pursue an offensive patent strategy, can use the indicator to measure the novelty of numerous published third-party patents and challenge their validity.