[The paper link is to be added]
In this work, we presents a total technology fitness landscape based on a technology embedding space and the estimated improvement rates of all domains in the space. The technology embedding space is trained via neural embedding techniques on both intrinsic (semantic) features and connective (citation) information to derive the high-dimensional embedding vectors for the 1,757 technology domains curated by Singh et al (2021) covering 95% of the patent database. The estimated improvement rates of these 1,757 domains are also drawn from Singh et al (2021). The technology fitness landscape exhibits a high hill related to information, electrical, and electronic technologies and a vast low plain of the rest domains. The construction of the technology fitness landscape based on neural embedding training enables a global picture and birds’ eye view of the co-evolution of heterogenous technology domains in the unified technology space.
The location of each domain is aligned to the 2D embedding map (the interactive figure above), and the color represents the rate. The heights correspond to the improvement rates of different domains.
Each item in the matrix represents the number of domains belong to the corresponding subcategories. The matrix has been normalized by column (the sum of each column equals to 1). This figure reveals a clear pattern of technological theme shifts from the global peak to the low plain.
To be added.
© 2013-2022 All Rights Reserved, SUTD Data-Driven Innovation Lab