Reproduce “Rankability-enhanced Revenue Uplift Modeling Framework for Online Marketing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bowei He, Yunpeng Weng, Xing Tang, Ziqiang Cui, Zexu Sun, Liang Chen, Xiuqiang He, Chen Ma
The RERUM framework, developed by City University of Hong Kong and Tencent FiT, enhances revenue uplift modeling by specifically addressing continuous, zero-inflated, and long-tail revenue distributions, while simultaneously improving the ranking accuracy of individuals by their predicted uplift. It achieved an average improvement of 20.61% in sales revenue LIFT@2 in online experiments at Tencent FiT, leading to an estimated gain of $430 million in assets under management monthly.
Key insights
- Accurate modeling of continuous, zero-inflated, and long-tail revenue distributions with ZILN loss is foundational for effective revenue uplift prediction and subsequent ranking.
- Theoretical analysis reveals that conventional MSE leads to looser error bounds for uplift ranking, demonstrating the need for explicit ranking optimization and tighter bounds through response ranking losses.
- Combining foundational response modeling with both implicit (response ranking) and explicit (listwise uplift ranking) objectives effectively addresses the dual challenge of predicting accurate revenue uplift values and correctly ranking individuals by their causal effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.