|
Abstract:
|
Pairwise interaction data, consisting of events involving pairs of actors over time, are now commonly available at the finest of temporal resolutions. While these data sets are often modeled in discrete-time, continuous-time models are needed to fully exploit the richness of information in the temporal structure. Existing continuous-time methodology utilize point processes to directly model interaction "contagion", whereby one interaction increases the propensity of future interactions among actors as dictated by a latent social network. In contrast, we argue that the consistency of interactions, rather than their frequency, is most indicative of a well-established underlying social relationship between actors. Different types of social relationships can be characterized by different interaction patterns with varying frequencies. To this end, we propose a novel temporal-relational point process model for continuous-time event data in which interactions are not assumed to be directly dependent on one another but are attributable to actor covariates and connections in the underlying network.
|
