HT-GNN: Hyper-Temporal Graph Neural Network for Customer Lifetime Value Prediction in Baidu Ads

TL;DR

HT-GNN tackles the challenge of predicting customer lifetime value ($LTV$) in advertising by jointly modeling demographic heterogeneity and dynamic, platform-induced behavioral uncertainty. It assembles a hypergraph-supervised learning module to capture inter-segment relationships, a Transformer-based temporal encoder with adaptive weighting to handle irregular sequences, and a task-adaptive mixture-of-experts with dynamic towers for multi-horizon forecasting, trained with a multi-loss objective that includes Jensen-Shannon divergence supervision. Empirical results on Baidu Ads data with $15\times 10^{6}$ users show robust improvements over strong baselines across $NRMSE$, $NMAE$, $AUC$, and $N$-GINI, especially for short-term horizons. The work advances advertising value prediction by integrating population structure and sequence dynamics, offering a framework potentially transferable to other domains with heterogeneous populations and evolving behavioral contexts.

Abstract

Lifetime value (LTV) prediction is crucial for news feed advertising, enabling platforms to optimize bidding and budget allocation for long-term revenue growth. However, it faces two major challenges: (1) demographic-based targeting creates segment-specific LTV distributions with large value variations across user groups; and (2) dynamic marketing strategies generate irregular behavioral sequences where engagement patterns evolve rapidly. We propose a Hyper-Temporal Graph Neural Network (HT-GNN), which jointly models demographic heterogeneity and temporal dynamics through three key components: (i) a hypergraph-supervised module capturing inter-segment relationships; (ii) a transformer-based temporal encoder with adaptive weighting; and (iii) a task-adaptive mixture-of-experts with dynamic prediction towers for multi-horizon LTV forecasting. Experiments on \textit{Baidu Ads} with 15 million users demonstrate that HT-GNN consistently outperforms state-of-the-art methods across all metrics and prediction horizons.

Figures (3)

• Figure 1: Illustration of uncertain behavior sequences under dynamic marketing strategies in advertising platforms
• Figure 2: Impact of Feature Similarity on LT and LTV Variance
• Figure 3: Hyper-Temporal Graph Neural Network Modeling Framework