Attribute-Enhanced Temporal Point Process for Personalized User Behavior Prediction

User-generated event sequence data is ubiquitous in real-world scenarios. Analyzing such data makes it possible to predict the user’s next event. Temporal point process (TPP) is a probabilistic method widely used to model the event sequence data. Existing TPP variants typically assume that the event sequences are independent and identically distributed (i.i.d.). However, such assumption is invalid in real-world scenarios, where the distributions of event sequences largely depend on specific users. Therefore, we propose an attribute-enhanced TPP method, which leverages user attribute information to assist in distinguishing the distributions of event sequences. Specifically, we consider user attributes’ dynamic influences on event sequences and static influences on events probability. We collect user event sequences and user attributes from two public datasets and one industrial dataset. The experiments on these three datasets demonstrate the importance of user attributes and the effectiveness of our method.