Identifying emerging technologies to envision a future innovation ecosystem: A machine learning approach to patent data.

Autor/inn/en	Choi, Youngjae; Park, Sanghyun; Lee, Sungjoo
Titel	Identifying emerging technologies to envision a future innovation ecosystem: A machine learning approach to patent data.
Quelle	In: Scientometrics, (2021) 7, S.5431-5476 PDF als Volltext Verfügbarkeit
Dokumenttyp	gedruckt; online; Zeitschriftenaufsatz
ISSN	0138-9130
DOI	10.1007/s11192-021-04001-1
Schlagwörter	Emerging technologies; Innovation ecosystem; Patent; Semi-supervised learning; Active learning
Abstract	Abstract With the rapid changes to technology as well as industry value chains, it has become essential for firms to identify emerging promising technologies that can better respond to external disruptive forces and be used to launch new businesses or improve current businesses. One of the most commonly used approaches in identifying emerging promising technologies is patent analysis. Patents have long been regarded as a useful source of data on technologies; accordingly, a number of previous studies have applied patents to define rising technologies. However, most previous studies have significantly relied on patent information in assessing promising technologies, whereas promisingness is determined by various other factors that are not explained by patent information. To overcome the limitation of previous approaches, this study proposes a hybrid approach considering both expert opinions and patent information to identify emerging promising technologies. For analysis, we firstly developed a set of criteria with which to evaluate potentially valuable patents, had experts evaluate only a portion of the patents from a larger patent portfolio of interest based on the criteria, and finally used the evaluation results to identify other potentially valuable patents from the rest of patents in the portfolio. Here, an active semi-supervised learning technique was applied, in which a small amount of labeled data (patents evaluated by experts) was used with a large amount of unlabeled data (the other patents from the portfolio). An analysis model consists of two layers—patents and patent attributes—with patent attributes such as technology characteristics used to classify patents into promising and unpromising ones. The proposed approach was applied to the automobile industry sector, and its usability was verified; the analysis results indicated that semi-supervised learning combined with active learning has potential in effectively searching for emerging promising technologies or filtering non-promising technologies with less human input. With only a small set of labeled patents, a large set of patents could be labeled, which saves time and effort when experts evaluate patents. Methodologically, this is an early attempt to introduce active semi-supervised learning in the context of patent analysis. Practically, the research findings enable expert opinions to be used effectively in identifying promising technologies and envisioning a future innovation ecosystem, making a balance between data- and expert-driven decision-making.
Erfasst von	OLC
Update	2023/2/05