Regularizing Treatment Leakage for LLM-based Uplift Estimation
Uplift modeling estimates the conditional average treatment effect (CATE) and supports industrial decisions such as promotional pricing. When large language models (LLMs) encode product text, treatment-revealing content can introduce shortcuts that do not transfer to deployment. We propose \emph{LLM-based Uplift Leakage Regularization} (\textsc{LULR}), which uses a CLUB-based penalty on the causal head's trainable shared representation while keeping the LLM encoder frozen. Controlled proxy experiments show a positive ranking-score interaction and higher scores than the tested weight-decay configuration; treatment-dependent evaluation limits their interpretation to exploratory model behavior. Structured-only comparisons provide reference performance, without establishing a general advantage over tree-based estimators. On a million-record industrial pricing RCT, \textsc{LULR} improves offline $\text{lift@}0.3$ in $4/5$ seeds, although the difference is not statistically significant and estimated policy values are near-tied. A separate one-week online A/B experiment records $5.45\%$ higher sales volume and $1.91\%$ higher GMV than the control arm. These aggregate results provide short-term business evidence for the complete deployed pricing policy; they do not isolate the regularizer's contribution.
Authors
- Xue Yuan (ORCID: https://orcid.org/0000-0002-0963-1998)
- Shuhan Wang (ORCID: https://orcid.org/0009-0006-6198-7031)
- Feng Xiao (ORCID: https://orcid.org/0009-0009-8103-3739)
- Shaomin Li (ORCID: https://orcid.org/0000-0002-0901-5124)
- Bo Feng (ORCID: https://orcid.org/0009-0005-0709-7903)
- Jinqiu Deng (ORCID: https://orcid.org/0009-0005-3282-5511)
- Zijiang Liu
- Yongzhi Qi
- Zihan Hong
- Shiwei Wang
- Wenqi Xiang
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.48550/arxiv.2611.20794
- Primary Topic
- Advanced Causal Inference Techniques
- Type
- preprint