A Bayesian-DLM-CF Framework for Real-Time Display Advertising

Click-through rate prediction underpins real-time bidding strategies in display advertising. We propose a unified approach that integrates beta-based Bayesian priors, Dynamic Linear Models, and collaborative filtering to address data sparsity, temporal dynamics, and neighbor relationships. A hierarchical Bayesian structure shares information across campaigns from the same advertiser, improving estimates when per-campaign data are limited. On a real-world dataset, our method outperforms baselines including standard collaborative filtering, random forest, and XGBoost, achieving superior log-loss and mean squared error.

Duke Scholars

Author David L. Banks Statistical Science