А. В. Горбунов As the global patent collection is widening, the complexity of patent documents search for assessing technique novelty, i. e. revealing the relevant art or prior art from public patent data, is increasing, too. Searching for this information, vast and complex, is challenging. Research findings evidence on the increasing scale of NLP use for more accurate and integrated patent search. Despite many achievements, the automated patent search system for appropriate accuracy and completeness has not been introduced. The author argues that development of new effective approaches to designing these systems is significantly limited due to the lack of the datasets ready for educating and testing. The automated acquisition of datasets of arbitrary configuration (with consideration for various selection criteria, i. e. documents by patent agency/agencies; all published documents for a limited period of time: document types; patent classification classes, etc.) would enable to eliminate limitations and build the datasets meeting the needs and goals set up by the systems designers. The author proposes new approaches to dataset acquisition, testing of automated art patent search systems, and assessment of these systems.
Научная статья 
патентная коллекция