CC-OR-Net: A Unified Framework for LTV Prediction through Structural Decoupling

TL;DR

The paper tackles Customer Lifetime Value (LTV) prediction under a zero-inflated long-tail distribution, introducing CC-OR-Net, a unified framework that structurally decouples ordinal ranking from regression via a cascaded ordinal decomposition, intra-bucket residual learning, and attention-guided high-value augmentation. It formalizes the ordinal problem with conditional exceedance probabilities $p_k$ to guarantee ordinal consistency and $O(K)$ complexity, augmented by a distillation module to align predictions with empirical distributions. The authors demonstrate superior trade-offs on three large industrial datasets, achieving improved Stratified Value Accuracy (SVA) and reduced high-value bias (AMBE) while maintaining competitive ranking metrics and efficient inference, including strong whale-recall performance. The work, validated through production deployments and cross-domain experiments, offers a principled approach to long-tail regression problems with critical minority segments and has potential applicability to other domains requiring robust ordinal ranking and high-value precision.

Abstract

Customer Lifetime Value (LTV) prediction, a central problem in modern marketing, is characterized by a unique zero-inflated and long-tail data distribution. This distribution presents two fundamental challenges: (1) the vast majority of low-to-medium value users numerically overwhelm the small but critically important segment of high-value "whale" users, and (2) significant value heterogeneity exists even within the low-to-medium value user base. Common approaches either rely on rigid statistical assumptions or attempt to decouple ranking and regression using ordered buckets; however, they often enforce ordinality through loss-based constraints rather than inherent architectural design, failing to balance global accuracy with high-value precision. To address this gap, we propose \textbf{C}onditional \textbf{C}ascaded \textbf{O}rdinal-\textbf{R}esidual Networks \textbf{(CC-OR-Net)}, a novel unified framework that achieves a more robust decoupling through \textbf{structural decomposition}, where ranking is architecturally guaranteed. CC-OR-Net integrates three specialized components: a \textit{structural ordinal decomposition module} for robust ranking, an \textit{intra-bucket residual module} for fine-grained regression, and a \textit{targeted high-value augmentation module} for precision on top-tier users. Evaluated on real-world datasets with over 300M users, CC-OR-Net achieves a superior trade-off across all key business metrics, outperforming state-of-the-art methods in creating a holistic and commercially valuable LTV prediction solution.

Figures (4)

• Figure 1: The challenging distribution of LTV, characterized by an Extreme Zero-Inflation peak, a severe Long-Tail, and a sparse but critical "Whale" User Segment.
• Figure 2: CC-OR-Net Overall Architecture: Unified framework processing heterogeneous features through four specialized modules: (left) shared representation learning encoding multi-modal inputs, (center) conditional cascaded ordinal decomposition with fixed quantile boundaries, (right-top) intra-bucket residual learning for fine-grained regression, and (right-bottom) high-value augmentation module with attention-guided augmentation. The bottom panel displays the characteristic long-tail distribution with highlighted high-value segments, while the data flow diagram shows prediction integration across modules.
• Figure 3: A 3D trade-off analysis on Domain 1. Axes represent overall accuracy (SVA) and high-value bias (AMBE), while bubble size indicates inference latency. The ideal region is top-right. CC-OR-Net demonstrates a superior balance, with the arrow showing the targeted improvement from our augmentation module.
• Figure 4: Whale finding efficiency on Domain 1, measured by Recall@5000. CC-OR-Net demonstrates a clear superiority in identifying the most valuable users within a fixed budget, a critical capability for maximizing marketing ROI.