CanniUplift Holistic Framework for Mitigating Seller and Incentive Cannibalization in E-commerce Uplift Modeling

Jul 16, 2026

Upload by Ezra

Personalized incentive allocation is essential in e-commerce, where uplift modeling estimates Individual Treatment Effects (ITE). Traditional approaches underperform in multi-seller settings that violate SUTVA by causing two forms of cannibalization: seller-level shifts of spend across shops without platform growth and incentive-level distortions from organic conversions and alternative rewards. CanniUplift unifies Platform-level Global Alignment (PGA) with Redemption-based Decomposition Denoising (RDD) and a Treat-Attention mechanism, improving incrementality estimation and platform ΔGMV via experiments and online deployment.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

Introduction

Abstract

CCS Concepts

Keywords

Method and Framework (PGA, RDD, Treat-Attention)

Experiments and Results

Ablation Studies

Deployment and Impact

CanniUplift: A Holistic Framework for Mitigating Seller and Incentive Cannibalization in E-commerce Uplift Modeling arXiv :2607 .05242v 1 [ cs .LG] 6 Jul 2026 Zuwang He∗ [zuwanghe.hzw@gmail.com](zuwanghe.hzw@gmail.com)[ ](zuwanghe.hzw@gmail.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Beijing, China Hanyu Gao [gaohanyu.ghy@alibaba-inc.com](gaohanyu.ghy@alibaba-inc.com)[ ](gaohanyu.ghy@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Beijing, China Xiangda Yan [xiangda.yxd@alibaba-inc.com](xiangda.yxd@alibaba-inc.com)[ ](xiangda.yxd@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Hangzhou, Zhejiang, China Shihao Shu∗ [shushihao.ssh@alibaba-inc.com](shushihao.ssh@alibaba-inc.com)[ ](shushihao.ssh@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Beijing, China Ziliang Zhang [zhangziliang.zzl@alibaba-inc.com](zhangziliang.zzl@alibaba-inc.com)[ ](zhangziliang.zzl@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Beijing, China Buyu Gao [gaobuyu.gby@alibaba-inc.com](gaobuyu.gby@alibaba-inc.com)[ ](gaobuyu.gby@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Hangzhou, Zhejiang, China Yuli Qu∗ [quyuli.qyl@alibaba-inc.com](quyuli.qyl@alibaba-inc.com)[ ](quyuli.qyl@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Hangzhou, Zhejiang, China Diwei Chen [chendiwei.cdw@alibaba-inc.com](chendiwei.cdw@alibaba-inc.com)[ ](chendiwei.cdw@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Hangzhou, Zhejiang, China Tanchao Zhu [tanchao.zhutc@alibaba-inc.com](tanchao.zhutc@alibaba-inc.com)[ ](tanchao.zhutc@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Hangzhou, Zhejiang, China Yumeng Li† [lym174806@alibaba-inc.com](lym174806@alibaba-inc.com)[ ](lym174806@alibaba-inc.com)[Taobao & Tmall Group of Alibaba](Taobao & Tmall Group of Alibaba)[ ](Taobao & Tmall Group of Alibaba)Beijing, China Junxiong Zhu [xike.zjx@alibaba-inc.com](xike.zjx@alibaba-inc.com)[ ](xike.zjx@alibaba-inc.com)Taobao & Tmall Group of Alibaba Hangzhou, Zhejiang, China Abstract Personalized incentive allocation is vital for e-commerce, where uplift modeling is the standard for estimating Individual Treatment Effects (ITE) . However, traditional models often fail in complex multiseller environments with violations ofthe Stable Unit Treatment Value Assumption (SUTVA) . We identify two critical challenges: Seller-level Cannibalization, where incentives shift expenditure between shops without growing the platform, and Incentive-level Cannibalization, where organic conversions or alternative rewards introduce significant noise into incrementality estimation. In this paper, we propose CanniUplift, a unified framework to mitigate these dual-source cannibalization effects. Specifically, we design Platform-level Global Alignment (PGA) to capture cross-shop substitution through global GMV consistency constraints. To tackle incentive-driven noise, we introduce Redemption-based Decomposition Denoising (RDD), which uses redemption behavior to decompose treated outcomes and reduce attribution noise within an entire-space framework. Furthermore, a Treat-Attention mechanism is designed to model intricate interactions between users’historical behaviors and current treatment options. Extensive experiments on both synthetic and large-scale industrial datasets demonstrate that CanniUplift significantly outperforms state-ofthe-art baselines. Ablation studies con

FAQs

What problems does CanniUplift address in e-commerce uplift modeling?

It addresses seller-level cannibalization, where incentives move spend between shops without growing platform GMV, and incentive-level cannibalization, where organic conversions or other rewards add noise to incrementality estimation.

How does Platform-level Global Alignment (PGA) reduce seller-level cannibalization?

PGA captures cross-shop substitution using global GMV consistency constraints, aligning treated outcomes across the platform level.

How does Redemption-based Decomposition Denoising (RDD) improve incentive-level noise handling?

RDD uses redemption behavior to decompose treated outcomes, reducing attribution noise within an entire-space framework and improving incrementality measures.

Jul 16, 2026

Upload by Ezra