Mapping Technology Space by Normalizing Patent Networks

Abstract

Technology is a complex system, with technologies relating to each other in a space that can be mapped as a network. The technology network's structure can reveal properties of technologies and of human behavior, if it can be mapped accurately. Technology networks have been made from patent data, using several measures of proximity. These measures, however, are influenced by factors of the patenting system that do not reflect technologies or their proximity. We introduce a method to precisely normalize out multiple impinging factors in patent data and extract the true signal of technological proximity, by comparing the empirical proximity measures with what they would be in random situations that remove the impinging factors. With this method, we created technology networks, using data from 3.9 million patents. After normalization, different measures of proximity became more correlated with each other, approaching a single dimension of technological proximity. The normalized technology networks were sparse, with few pairs of technology domains being significantly related. The normalized network corresponded with human behavior: we analyzed the patenting histories of 2.8 million inventors and found they were more likely to invent in two different technology domains if the pair was closely related in the technology network. We also analyzed 250 thousand firms' patents and found that, in contrast, firms' inventive activities were only modestly associated with the technology network; firms' portfolios combined pairs of technology domains about twice as often as inventors. These results suggest that controlling for impinging factors provides meaningful measures of technological proximity for patent-based mapping of the technology space, and that this map can be used to aid in technology innovation planning and management.

arXiv comment13 pages + 23 pages Appendix and SI

References in corpus (8)

• Fast unfolding of communities in large networks
• Finding and evaluating community structure in networks
• Vertex similarity in networks
• A tool for filtering information in complex systems
• Mapping Topics and Topic Bursts in PNAS
• Topology and Evolution of Technology Innovation Networks
• Faster unfolding of communities: speeding up the Louvain algorithm
• Measuring Technological Distance for Patent Mapping

Cited by in corpus (6)

• TechNet: Technology Semantic Network Based on Patent Data
• Guiding Data-Driven Design Ideation by Knowledge Distance
• Evolution and structure of technological systems - An innovation output network
• Long-run dynamics of the U.S. patent classification system
• Multilayer patent citation networks: A comprehensive analytical framework for studying explicit technological relationships
• Innovative activities of Activision Blizzard: A patent network analysis