Revealing Structural Patterns of Patent Citation by a Two-Boundary Network Model Based on USPTO Data

The patent is one of the carriers of scientific research and development, and an indicator of technical innovation. As a promising approach for modeling complex systems, complex networks could provide the sound theoretical framework for developing proper simulation models. Many researchers use the relations of patent citations and transfers to study knowledge propagate and output in the network. However, knowledge flow in patents should be fully considered by substantial and fruitful connections, both in the process of knowledge application and knowledge output. In this paper, we present a two-boundary network model with knowledge application boundary and knowledge output boundary to reveal the patent citation patterns in knowledge flow. The feasibility and effectiveness of the two-boundary network model are proved with theory and experiment. Utilizing 578,678 patents from the United States Patent and Trademark Office between 2015 and 2018 with the two-boundary network model, we put up a fixed effect ordinary least square equation to reveal the patent impacts of different structural patterns. Experimental results show that, in the perspective of structural patterns, the highly impacted patents without assigning are greatly influenced by their scientific literature references and provide knowledge for other assigned patents. However, considering all the fixed effect factors, patents that transfer knowledge from other patents to assigned patents are more likely to become highly impacted patents. Besides, we find the two-boundary network model fits the real patent knowledge flow well by comparing it with the other models.