CanniUplift addresses two forms of cannibalization that can distort e-commerce uplift modeling: incentives shifting spending across sellers without increasing platform GMV, and organic conversions or alternative rewards creating attribution noise. The framework combines Platform-level Global Alignment (PGA), Redemption-based Decomposition Denoising (RDD), and a Treat-Attention mechanism for treatment-user interactions. According to the abstract, experiments on synthetic and large-scale industrial datasets improved wAUUC and wQINI over state-of-the-art baselines. An online deployment reported a 4.08% relative increase in platform-wide incremental GMV over the production baseline, alongside improved ROI in online A/B tests.
No heat snapshots are available in the last 24 hours.