Measuring Growth Potential of Communities in the Patent Citation Network

The United States Patent and Trademark Office (USPTO) requires inventors who submit patents to include citations to preexisting patents containing similar solutions to the technical problem the patent is solving. This has resulted in a massive network of connected technological ideas spanning decades, and serves as an invaluable proxy to measure and model technological innovation [3]. Such investigations have practical value, offering insights, for example, into the relative importance of an individual patent, how certain factors such as patent age and inventor predict patent importance, and the emergence and evolution of new technological fields. These insights can in turn be used to determine the economic value of patent portfolios and drive research and development decisions by firms. Modeling the patent citation network as a graph allows us to uncover insights on the processes of technological innovation using graph properties alone. Node degree, for example, can be used in evaluating a patent’s relative importance and predicting its success in terms of the number of citations it receives; the out degree. In our preliminary analysis, we hypothesized that some of the predictors of a patent’s importance might include both inand outdegrees, age, how similar a patent is to other patents, and some measure of how ”original” a patent is based on its citations to communities not containing or related to that patent. Finally, community detection and tracking gives us a precise method of determining technological fields and their evolution. In this paper we will focus on detecting communities within the patent network and tracking their growth over time, then evaluating which network properties predict long-term growth of the communities. The paper is divided into four sections: a review of existing literature on the subjects of community detection and tracking and prior investigations into citation networks; our procedure, including data definition and cleansing and the models, algorithms, and techniques used; our results; and finally, a discussion and suggestions for further research.