Score-Based Learning of Graphical Event Models with Background Knowledge Augmentation

Graphical event models (GEMs) are representations of temporal point process dynamics between different types. Many real-world applications however involve limited stream data, making it challenging to learn GEMs from data alone. In this paper, we introduce approaches that can work together in a score-based learning paradigm, augment with potentially types background knowledge. We propose novel scores for an important parametric class GEMs; particular, Bayesian score leveraging prior information as well more practical simplification involves fewer parameters, analogous networks. also framework incorporating easily assessed qualitative knowledge domain experts, the form statements such `event X depends on Y' or Y makes likely'. The proposed has interpretations and be deployed by any learner. Through extensive empirical investigation, demonstrate benefits augmentation while low-data regime.