Machine learning-based method to cluster a converging technology system: The case of printed electronics

Abstract

Technology convergence is considered one of the cornerstones of technological innovation as a phenomenon emerging at the intersection of two previously unrelated fields of technology. The new technological system is a new combination of knowledge types, technology components and intersections. For this matter, analyzing patents is an essential part for strategic decision making. However, the manual analysis of large amounts of patent semantics is often time-consuming, extensive, and difficult even for experts. To enhance manual patent analyses, new machine learning-based techniques are gaining increasing interest. This study aims to enrich this methodological research by developing and evaluating an unsupervised text-mining approach to automatically cluster patents of two knowledge types into four technology components. To this end, this study presents a five-step method including the comparison between different algorithms and design choices. This method is applied to printed electronics-relevant patents extracted from the Derwent World Patent Index and enables to draw recommendations for automated patent analyses. The findings show different significances for types of components: while components of the specialized knowledge type could be predicted with significance, components of the design knowledge types could not provide significant results.