Type:

Year:

Author:

Supervisor:

Institution:

Page(s):

Website:

Abstract:

Developing new technologies is one of the most important goals of contemporary scientific and industrial research. Understanding how a technology domain evolves and its current state is invaluable in an ecosystem seeing the speed of technology evolution increasing at rapid pace. An overview of existing literature showed that while there is a significant volume of research focusing on using patents to study technology change, most of this research, in a technology evolution context, focuses on studying technology trajectories and convergence, with limited research combining insights from research based on other resources (i.e. paper citations) and applied to patent networks. Moreover, the review of literature shows that the majority of patent analysis methods focus on exploring technology development trajectories by examining the direct citations of patents. While this approach provides insight into the generational flow of knowledge, it provides little insight into how existing patents might combine and co-contribute to a future patent in the form of co-citations. Finally, a review of literature showed that the vast majority of patent-based methods for life cycle analysis as well as prediction base themselves on models derivate of the basic S – Curve model, providing little understanding of the underlying dynamics of patent attributes and their correlation to the life cycle phases. Quantitative methods not based on the S-Curve model mostly do not use patent data as a primary data source and give limited insight into the future knowledge flows.
This thesis aims to present a novel way of exploring the life cycle stages of a technology domain by conducting a dynamic growth analysis of a patent citation network, with patents being used as proxies for technological invention and patent citations representing the flow of knowledge. Additionally, new insights into the dynamics of the flow of knowledge within a technology domain are made by applying several link prediction algorithms to patent co-citation networks with the goal of identifying the link prediction algorithms most successful in describing the underlying intuition of co-citation network growth. Moreover, the dynamics of co-citation creation are explored by determining which part of a technology domains life cycle influences the link prediction algorithms precision the most and when the predicted links occur.
Two technology domains are explored; the car headlights technology domain representing a mature technology and the neuromorphic hardware technology domain representing an emerging technology. The choice of two technologies different in nature is deliberate; this way, the evolution of two different types of technologies can be explored and compared, helping to identify the particularities of each technologies evolution. The presented methodology for exploring the evolution of a technology domain consists of creating a dataset containing patents representing the studied technology domain and conducting a pair of technology life cycle (TLC) analyses. The first life cycle analysis is performed using an established method based on the cumulative number of patent applications over time, while the second is performed using a novel method based on analysing the dynamic growth of a patent citation network. An algorithm is introduced to convert the patent citation network into a patent co-citation network. Several link prediction algorithms are applied to the created co-citation networks to explore the underlying intuition governing co-citation network growth.
The study results show that a correlation exists between the stages of a technology domains life cycle and changes in the dynamics of patent citation network growth. The transition of the mature technology domains life cycle stage from growth to maturation correlates with a noticeable change in patent citation growth dynamics. Additionally, examining the emerging technology domain, it is found that a correlation exists between the time when an exponential increase in the number of inventions starts and a change in the dynamic of patent citation network growth. The Preferential Attachment link prediction algorithm is shown to be the most successful in predicting missing links in a mature technology. The results indicate that the patent co-citation occurring at the end of the growth TLC stage and the start of the maturation TLC stage contribute the most to the algorithm's precision. Moreover, it is demonstrated that most of the predicted missing links occur in a time frame closely following the application of the link-prediction algorithm. The results of studying the emerging technology domain show that link prediction algorithms have a significantly lower success in predicting missing links. The Adamic/Adar, Resource Allocation Index and Jaccard Coefficient show a moderate to low success in predicting missing links while the Preferential Attachment shows no precision.
This thesis provides a contribution in both a theoretical and practical context. In a theoretical context, the theoretical background of previous studies is expanded with new insights in patent citation networks growth as well as the dynamics of patent co-citation network growth. In a practical context, it is demonstrated that companies involved in planning for the short term should consider the knowledge contained in the patents relevant to their respective fields, reinforcing the notion that proper knowledge management is an invaluable tool to companies aspiring to innovate or produce innovative products.

Keywords: