Engineering Knowledge Graph From Patent Database

Abstract

We propose a large scalable engineering knowledge base as an integrated knowledge graph, comprising sets of (entity, relationship, entity) triples that are real-world engineering ‘facts’ found in the patent database. We apply a set of rules based on the syntactic and lexical properties of claims in a patent document to extract entities and their associated relationships that are supposedly meaningful from an engineering design perspective. Such a knowledge base is expected to support inferencing, reasoning, recalling in various engineering design tasks. The knowledge base has a greater size and coverage in comparison with the previously used knowledge bases in the engineering design literature.

Patent Data for Engineering Design: A Review

“…A few studies adopted standard text pre-processing pipelines, including lemmatization, stemming and stop-words removal techniques to clean the raw design text for further processing. In addition, design knowledge is extracted in certain forms of templates from given patent text utilizing syntactic dependencies to support the automation of TRIZ (Yamamoto et al, 2010) and construction of knowledge graphs (Fantoni et al, 2013;Siddharth et al, 2022). Besides, topic modelling algorithms such as non-negative matrix factorization (Song et al, 2020;Song and Fu, 2019) and latent semantic analysis have been applied on patent texts to represent design repositories in a more structured form.…”

“…Several early studies leveraged pre-trained common-sense semantic networks (Linsey et al, 2012) or manually curated ontology-based knowledge graphs (Atherton et al, 2018;Hagedorn et al, 2015;Jiang et al, 2018) to support design innovation and problem solving. Until recently, the patent database is mined to construct large-scale cross-domain engineering semantic networks and knowledge graphs (Sarica et al, 2020;Siddharth et al, 2022), serving as a knowledge infrastructure to support data-driven engineering design research and practice. Artificial neural networks and deep learning, because of their abilities to learn complex patterns from big data, have been employed in the patent-for-design literature.…”

“…TechNet has been used for design representation , prior art retrieval (Sarica et al, 2019), idea generation , and concept evaluation (Han et al, 2020). Siddharth et al (2022) created a large and scalable engineering knowledge graph based on the entire USPTO database, which can be used to support inference, reasoning, and knowledge exploration in engineering design applications.…”

“…Moreover, every patent is assigned to class(es) by patent examiners, making patent data ready for supervised machine learning applications. In recent years, several design research groups actively explored cutting-edge data science techniques to mine the patent database for diverse applications, such as design representation, design space exploration, design prior art searching, stimuli recommendation, idea generation and evaluation, etc (Jiang et al, 2021;Liu et al, 2020;Luo et al, 2021;Siddharth et al, 2022;Song et al, 2020;Song and Fu, 2019). These patent-for-design studies relied on a broad collection of methods, ranging from classic statistical analysis to latest network analysis, natural language processing (NLP), machine learning and data visualizations.…”

Natural language processing in-and-for design research

“…Scholars have indicated the possibility of extracting triples from the patent text (Soo et al 2006;Cascini and Zini 2008;Korobkin et al 2015). Siddharth et al (2021), for example, apply some rules to extract facts from patent claims by exploiting the syntactic and lexical properties. While patents could offer rule-based extraction methods due to consistent language, scientific articles require a mix of rule-based, ontology-based and supervised approaches.…”