Abstract

Patent citation analysis is considered a useful tool for technology impact analysis. However, the outcomes of previous methods do not provide a fair reflection of a technology’s future prospects since they are based on deterministic approaches, assuming that future trends will remain the same as those in the past. As a remedy, we propose a Hawkes process-based patent citation analysis method to assess the future technological impact and uncertainty of a technology in a time period of interest by employing the future citation counts of the relevant patents as a quantitative proxy. For this, we construct a citation interval matrix from the United States Patent and Trademark Office (USPTO) database, and employ a Hawkes process - a special case of path-dependent stochastic processes - as a method for patent citation forecasting. Specifically, the Hawkes process models the idiosyncratic and dynamic behaviours of a technology’s evolution and obsolescence by increasing the likelihood of another subsequent citation by oneself (i.e., self-excitation) and decaying the likelihood back towards the initial level naturally. A case study of the patents about molecular amplification diagnosis technology shows that our method outperforms previous deterministic approaches in terms of accuracy and practicality.