Predictive Modeling of Asynchronous Event Sequence Data

Large volumes of temporal event data, such as online check-ins and electronic records of hospital admissions, are becoming increasingly available in a wide variety of applications including healthcare analytics, smart cities, and social network analysis. Those temporal events are often asynchronous, interdependent, and exhibiting self-exciting properties. For example, in the patient\u27s diagnosis events, the elevated risk exists for a patient that has been recently at risk. Machine learning that leverages event sequence data can improve the prediction accuracy of future events and provide valuable services. For example, in e-commerce and network traffic diagnosis, the analysis of user activities can be used to predict and control dynamic user traffic demand to improves risk response efficiency. In this work, we investigate and design novel point process-based models and learning algorithms to analyze dynamic event sequence data from various aspects. (1) We first propose local low-rank Hawkes processes to capture the mutual influences between sequences of multiple event types. (2) We then develop geometric Hawkes processes to integrate geometric structures to point processes based on graph convolutional recurrent neural networks to improve prediction accuracy. (3) We introduce several novel fairness metrics to penalize the event likelihood function in order to tackle the challenge of data bias and the amplified through self-excitation in point processes. (4) We also propose a novel list-wise fairness criterion for point processes that can efficiently evaluate the ranking fairness in event prediction, and present a strict definition of the unfairness consistency property of a fairness metric