Cross-Head Attention Uplift Network with Inverse Propensity Score under Unobserved Confounding

2026-06-25

AI Summary

This study addresses the challenge of balancing between-group similarity exploitation and debiasing in uplift modeling under unobserved confounding. To this end, the authors propose the Cross-Attention Uplift Network (CHAUN) and a Robust Adversarial Inverse Propensity Score (RA-IPS) method. CHAUN enhances inter-group representation learning by integrating treatment and control group features through shared embeddings and a cross-attention mechanism. RA-IPS mitigates confounding bias by adversarially learning robust weighting without access to the true propensity scores. Theoretically, the work establishes, for the first time, that individual treatment effects remain identifiable even in the presence of unobserved confounders, provided the true propensity scores are known. Empirical results demonstrate that the proposed approach achieves up to a 25.6% improvement in QINI score on CRITEO-UPLIFT, LAZADA, and e-commerce datasets, with RA-IPS outperforming standard IPS by 5.4% under unobserved confounding.

Abstract

Uplift modeling, crucial for estimating individual treatment effects (ITE), faces dual challenges: flexibly leveraging inter-group similarity to enhance discriminative power and debiasing under unobserved confounding scenarios. In this paper, we propose the Cross-Head Attention Uplift Network (CHAUN) and Robust Adversarial Inverse Propensity Score (RA-IPS) method to address these limitations. CHAUN employs shared feature embeddings and cross-head attention mechanisms to dynamically integrate treatment-specific and control-specific representations, enhancing inter-group correlation modeling. Theoretically, we prove that access to the true propensity scores ensures ITE identifiability even with unobserved confounders. For practical scenarios lacking true propensity scores, RA-IPS adversarially optimizes propensity weights within constrained uncertainty sets to mitigate bias from unobserved variables. Experiments on public datasets (CRITEO-UPLIFT, LAZADA) and a production e-commerce dataset demonstrate CHAUN's superiority over state-of-the-art uplift models, achieving relative improvements of up to 25.6% in QINI scores. RA-IPS further enhances robustness, outperforming standard IPS by 5.4% under unobserved confounding. The results validate the effectiveness of our proposed methods in real-world causal inference tasks.

Problem

• uplift modeling
• individual treatment effect
• unobserved confounding
• inter-group similarity
• bias correction

Innovation

• Cross-Head Attention
• Uplift Modeling
• Inverse Propensity Score
• Unobserved Confounding
• Individual Treatment Effect