Problem
Research questions and friction points this paper is trying to address.
auto-bidding
sliding-window evaluation
long-term advertising effectiveness
cross-episode coupling
efficiency constraints
Beyond Single-Episode Optimization: Sliding-Window Aware Generative Auto-Bidding for Long-Term Advertising Effectiveness
🤖 AI Summary
This work addresses the limitations of traditional automated bidding methods, which treat each day as an isolated unit and struggle with sparse conversions and cross-day coupling under sliding-window evaluation. To overcome these challenges, the authors propose SWAG-Bid, a novel framework that introduces, for the first time, a sliding-window-aware mechanism to optimize long-term advertising performance through a hierarchical architecture. The upper-level planner generates compatible candidate bidding schedules via Multi-Window Model Predictive Sampling (MWMS), while the lower-level controller dynamically adjusts bids using a state-adaptive gating mechanism (PSG-AdaLN) to balance budget adherence and constraint satisfaction. Integrated with a Masked Trajectory Model and dedicated Return/Cost-to-Go channels, SWAG-Bid significantly improves both constraint compliance and value acquisition under sliding-window evaluation, as demonstrated in experiments on AuctionNet-Sparse and live A/B tests on AliExpress.
📝 Abstract
Auto-bidding systems optimize bids to maximize value under efficiency constraints such as Cost-Per-Action (CPA). Existing methods treat each day as an independent episode. However, many advertisers produce value so sparsely that per-day efficiency ratios become statistically unreliable, undermining advertiser retention. Platforms therefore evaluate window-level efficiency over sliding windows of $W{=}7$ days, ensuring fair evaluation and long-term advertising effectiveness. This creates cross-episode coupling: each day's bidding decisions affect up to $W$ overlapping windows, so setting daily targets requires anticipating future market conditions. We propose SWAG-Bid (Sliding-Window Aware Generative Auto-Bidding), a hierarchical framework decomposing this challenge into episode-level planning and step-level execution. The planner uses a Masked Trajectory Model to forecast markets and generate candidate plans, scored across all overlapping windows by Multi-Window Model Predictive Control Sampling (MWMS) with exponential confidence decay. The controller adjusts reliance on this guidance through a state-adaptive gate, Per-Step Gated Adaptive Layer Normalization (PSG-AdaLN), complemented by Return-to-Go and Cost-to-Go channels carrying budget and constraint information. Experiments on AuctionNet-Sparse and online A/B tests on AliExpress show that SWAG-Bid achieves competitive constraint satisfaction and value acquisition under sliding-window evaluation.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Sliding-Window Awareness
Generative Auto-Bidding
Multi-Window Model Predictive Control
Cross-Episode Coupling
State-Adaptive Gating
🔎 Similar Papers
No similar papers found.
