A Stochastic Gradient Relational Event Additive Model for modelling US patent citations from 1976 until 2022

Abstract

Until 2022, the US patent citation network contained almost 10 million patents and over 100 million citations. To overcome limitations in analyzing such complex networks, we propose a stochastic gradient relational event additive model (STREAM) that models the relationships between citing patents as events that occur over time, where predictors are modeled through B-splines. Our model identifies key factors driving patent citation and reveals insights, such as time windows where citations are more likely and the relevance of the increasing citation numbers per patent. Overall, the STREAM offers the potential for capturing dynamics in large sparse networks while maintaining interpretability.