Event prediction using vector representation in hawkes processes

Abstract

To predict future events, previous research models diffusion patterns of events. Limitation of these research is that they could only model fixed number of event types. For this reason, existing models could not predict new types of events. To solve this problem, we introduce Vectorized Hawkes Process, which extends Hawkes process in continuous vector space. We conduct experiment on synthetic data and two real world datasets. Our model’s predictive power and estimate power of relations between events are demonstrated by the experimental results.