Rankability-enhanced Revenue Uplift Modeling Framework for Online Marketing

TL;DR

The paper tackles revenue uplift modeling in online marketing, focusing on continuous long-tail responses and uplift ranking. It introduces the RERUM framework, which integrates a zero-inflated lognormal (ZILN) regression loss, two tight response-ranking bounds (within-group and cross-group), and a listwise uplift ranking objective to optimize population-wide rankability. Theoretical analysis and extensive offline and online experiments show that RERUM improves uplift ranking (AUUC, AUQC, KRCC, LIFT@30) and delivers meaningful online revenue gains on a real-world Tencent FiT platform. The work demonstrates significant practical impact for marketing ROI by directly targeting the ranking quality of uplift estimates and revenue-driven decisions.

Abstract

Uplift modeling has been widely employed in online marketing by predicting the response difference between the treatment and control groups, so as to identify the sensitive individuals toward interventions like coupons or discounts. Compared with traditional \textit{conversion uplift modeling}, \textit{revenue uplift modeling} exhibits higher potential due to its direct connection with the corporate income. However, previous works can hardly handle the continuous long-tail response distribution in revenue uplift modeling. Moreover, they have neglected to optimize the uplift ranking among different individuals, which is actually the core of uplift modeling. To address such issues, in this paper, we first utilize the zero-inflated lognormal (ZILN) loss to regress the responses and customize the corresponding modeling network, which can be adapted to different existing uplift models. Then, we study the ranking-related uplift modeling error from the theoretical perspective and propose two tighter error bounds as the additional loss terms to the conventional response regression loss. Finally, we directly model the uplift ranking error for the entire population with a listwise uplift ranking loss. The experiment results on offline public and industrial datasets validate the effectiveness of our method for revenue uplift modeling. Furthermore, we conduct large-scale experiments on a prominent online fintech marketing platform, Tencent FiT, which further demonstrates the superiority of our method in real-world applications.

Figures (4)

• Figure 1: The histogram for the revenue frequency distribution of the A/B test in our deployment platform's online fintech application scenario. Left: raw revenue values; Right: revenue values after the logarithmic transformation.
• Figure 2: The overall framework of uplift estimation model. The modules in the dotted rectangular boxes are optional according to the specific base model selection. The original versions of base models take MSE as the response regression loss.
• Figure 3: The hyperparameter sensitivity analysis results. The line chart corresponds to the AUUC metric and the column chart corresponds to the LIFT@30 metric.
• Figure 4: Overview of the online deployment platform.