Three Real-World Datasets and Neural Computational Models for Classification Tasks in Patent Landscaping

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11498 - 11513 December 7-11, 2022 ©2022 Association for Computational Linguistics Three Real-World Datasets and Neural Computational Models for Classification Tasks in Patent Landscaping Subhash Chandra Pujari1,3 Jannik Strötgen1 Mark Giereth2 Michael Gertz3 Annemarie Friedrich1 1Bosch Center for Artificial Intelligence, Renningen, Germany 2Robert Bosch GmbH, Stuttgart, Germany 3Institute of Computer Science, Heidelberg University, Heidelberg, Germany firstname.lastname@de.bosch.com gertz@informatik.uni-heidelberg.de Abstract Patent Landscaping, one of the central tasks of intellectual property management, includes se- lecting and grouping patents according to user- defined technical or application-oriented cri- teria. While recent transformer-based models have been shown to be effective for classifying patents into taxonomies such as CPC or IPC, there is yet little research on how to support real-world Patent Landscape Studies (PLSs) us- ing natural language processing methods. With this paper, we release three labeled datasets for PLS-oriented classification tasks covering two diverse domains. We provide a qualitative analysis and report detailed corpus statistics. Most research on neural models for patents has been restricted to leveraging titles and abstracts. We compare strong neural and non-neural base- lines, proposing a novel model that takes into account textual information from the patents’ full texts as well as embeddings created based on the patents’ CPC labels. We find that for PLS-oriented classification tasks, going beyond title and abstract is crucial, CPC labels are an effective source of information, and combining all features yields the best results. 1 Introduction A patent is a public document granting the exclu- sive rights to an invention, e.g., a product or a process that provides a new technical solution to a problem. When entering new markets or devel- oping new products, it is of utmost importance for organizations such as companies to be aware of the patent landscape, i.e., the existing patents with regard to their business endeavor in order to ensure their freedom to operate. Experiments conducted with expert patent examiners in a context of a fea- sibility study on prior art search showed that while fully automating the process is infeasible, Natu- ral Language Processing (NLP) methods can help to significantly reduce time and cost by assisting patent examiners (Setchi and Spasic, 2020). title abstract claims description CPC labels SciBERT classifier Port Fuel Injector Fuel Rail Water Injection Dual Injection Bi-Fuel Injector ... Figure 1: A Patent Landscape Study commonly in- volves the classification of patents into a set of business- or application-oriented target classes. The task of obtaining an overview of the rele- vant intellectual property with the aim of support- ing strategic decisions is also called performing a Patent Landscape Study (PLS). A PLS consists of three steps: search, classification, and analysis. The first step identifies relevant documents that are then grouped into a set of user-defined categories. Finally, the labeled dataset is used to derive essen- tial insights. Most efforts towards automating the PLS process (Abood and Feltenberger, 2018; Choi et al., 2022) focus on the first step of the process. In contrast, in …

Content truncated to 500 words