Abstract
Recent technological advances have made it easier to collect large andcomplex networks of time-stamped relational events connecting two or moreentities. Relational hyper-event models (RHEMs) aim to explain the dynamics ofthese events by modeling the event rate as a function of statistics based onpast history and external information. However, despite the complexity of the data, most current RHEM approachesstill rely on a linearity assumption to model this relationship. In this work,we address this limitation by introducing a more flexible model that allows theeffects of statistics to vary non-linearly and over time. While time-varyingand non-linear effects have been used in relational event modeling, we takethis further by modeling joint time-varying and non-linear effects using tensorproduct smooths. We validate our methodology on both synthetic and empirical data. Inparticular, we use RHEMs to study how patterns of scientific collaboration andimpact evolve over time. Our approach provides deeper insights into the dynamicfactors driving relational hyper-events, allowing us to evaluate potentialnon-monotonic patterns that cannot be identified using linear models.