Doubly Robust Estimation of Causal Effect on CVR with Targeted Regularization
The paper argues that CVR causal estimates need an estimator built for click-to-conversion chains, not just debiased loss.
The authors introduce a doubly robust estimator for post-click conversion-rate effects that accounts for selection from clicked samples. They say it has stronger convergence behavior when nuisance components are estimated with flexible models such as neural networks. A targeted-regularization framework is added to improve stability in practice. Experiments on synthetic and real-world data are reported as more robust than naive combinations of loss debiasing and standard causal estimators. ArXiv · AI/CL/LG's note
The authors introduce a doubly robust estimator for post-click conversion-rate effects that accounts for selection from clicked samples. They say it has stronger convergence behavior when nuisance components are estimated with flexible models such as neural networks. A targeted-regularization framework is added to improve stability in practice. Experiments on synthetic and real-world data are reported as more robust than naive combinations of loss debiasing and standard causal estimators. ArXiv · AI/CL/LG's note
score 4