C-NTPP: Learning Cluster-Aware Neural Temporal Point Process

Event sequences in continuous time space are ubiquitous across applications and have been intensively studied with both classic temporal point process (TPP) its recent deep network variants. This work is motivated by an observation that many of event data exhibit inherent clustering patterns terms the sparse correlation among events, while such characteristics seldom explicitly considered existing neural TPP models whereby history encoders often embodied RNNs or Transformers. In this work, we propose a c-NTPP (Cluster-Aware Neural Temporal Point Process) model, which leverages sequential variational autoencoder framework to infer latent cluster each belongs sequence. Specially, novel event-clustered attention mechanism devised learn then aggregate them together obtain final representation for event. Extensive experiments show achieves superior performance on real-world synthetic datasets, it can also uncover underlying correlations.