Automatic extraction of inventive information out of patent texts in support of manufacturing design studies using Natural Languages Processing

Intelligent manufacturing systems are constantly evolving in diversity and complexity. The rise of numeric era, ruled by the keywords industry 4.0 or industry of the future imposes to companies to invent new processes and solve an ever increasing quantity of problems. Paradoxically, even if techniques of inventive problem-solving progress in diversity, their ability to face this world-wide challenge do not grow accordingly. However, thanks to Natural Languages Processing (NLP), actors of invention can now count on information contents as an assistant through its textual data. Patent texts are of particular interest since they are an important and constantly renewed source of inventive information. This situation leads to the difficulty, for scientists and engineers, to permanently manage new masses of information from recent domains well beyond their reading capacity. Our research, based on the combination of the theory of inventive problem-solving (also known as TRIZ) and NLP, has made it possible to extract quickly and in a relevant way from patent texts, concepts that contain information useful for formulating an inventive problem. In this paper, we present our methodology for the automatic extraction of inventive information from patent texts and measure our technique to a classical human-led information gathering. Our results show a significant reduction of experts time solicitation, for an increase of 36% in the extraction of useful information. A case study applied to microplastics harvesting from the ocean illustrates our point.