Identifying the technology convergence using patent text information: A graph convolutional networks (GCN)-based approach

• Published in: Technological forecasting & social change 2022-03, Vol.176, p.121477, Article 121477
• Main Authors: Zhu, Chen, Motohashi, Kazuyuki
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Artificial neural networks; Case studies; Classification; Consumer goods; Convergence; Distributed ledger; Distributed ledger technology; Graph convolution networks; Indicators; Patent analysis; Technology; Technology convergence; Technology utilization
• ISSN: 0040-1625; 1873-5509
• DOI: 10.1016/j.techfore.2022.121477

•A new semantic method to monitor technology convergence using a graph convolutional network (GCN)-based graph autoencoder (GCN-based GAE) is proposed.•The proposed method outperforms existing studies using information regarding cross-citations and co-occurrence of international patent classification classes.•Potential indicator development based on the proposed method is discussed. The potential for new values and products created by technology convergence to disruptively transform existing industries and markets is high. In this regard, it has been crucial for companies to understand and identify potential convergence patterns as early as possible to make timely strategic plans. This study proposes a new semantic method by showing how a graph convolutional network model can be used to monitor technology convergence. In particular, the model is trained to generate patents and technology keyword vectors from which new indicators are derived. We validate these new indicators and show that the proposed method outperforms existing studies using information regarding cross-citations and co-occurrence of international patent classification classes. Furthermore, we presented the usefulness of the proposed method to monitor technology convergence using a case study of the convergence between artificial intelligence (AI) and distributed ledger technology (DLT). The results show that convergence between AI and DLT is driven mainly by employing AI for DLT, and the role of each keyword (sub-domain) in the convergence process is also presented.