A Point Process Model for Optimizing Repeated Personalized Action Delivery to Users

TL;DR

Addresses causal inference for repeated personalized action delivery in interactive systems within a Bayesian decision framework. Models event histories as temporal marked point processes with a neural parameterization $P_{\phi}$ and a policy $\pi_\xi$ to optimize actions via the predictive distribution and expected utility $\mathcal{U}(\xi; \mathcal{D}, S)$. Learns from logs by maximizing the posterior predictive likelihood $P(\theta|\mathcal{D})$ and performing gradient-based optimization. Contributions include a general formalism for repeated interventions, tractable event-likelihood factorization, and neural TPP implementations enabling sampling and scalable policy optimization. The framework is applicable to online settings such as real-time bidding and recommender systems, offering principled personalized action delivery under stationarity and SUTVA.

Abstract

This paper provides a formalism for an important class of causal inference problems inspired by user-advertiser interaction in online advertiser. Then this formalism is specialized to an extension of temporal marked point processes and the neural point processes are suggested as practical solutions to some interesting special cases.