Multi-agent Auto-Bidding with Latent Graph Diffusion Models

2025-03-04

AI Summary

This paper addresses KPI-constrained bidding optimization for large-scale online advertising in multi-agent automated bidding scenarios under uncertain, sparse, and stochastic competitive environments. We propose the first framework integrating learnable graph embeddings with a planning-oriented latent diffusion model (LDM), jointly modeling exposure opportunity dependencies and multi-agent auction dynamics while enabling constraint-aware bid trajectory generation. Our method unifies graph neural networks, LDMs, reward-aligned posterior fine-tuning, multi-agent reinforcement learning, and KPI-constrained optimization. Evaluated on both real-world and synthetic auction environments, it significantly improves ROI, GMV, and budget utilization rate, while enhancing prediction accuracy of auction outcomes. The framework establishes a novel paradigm for KPI-driven, multi-agent bidding—advancing beyond conventional single-agent or heuristic approaches by explicitly capturing inter-agent competition structure and hard KPI constraints within a generative planning framework.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Online Learning & Bandits
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This paper proposes a diffusion-based auto-bidding framework that leverages graph representations to model large-scale auction environments. In such settings, agents must dynamically optimize bidding strategies under constraints defined by key performance indicator (KPI) metrics, all while operating in competitive environments characterized by uncertain, sparse, and stochastic variables. To address these challenges, we introduce a novel approach combining learnable graph-based embeddings with a planning-based latent diffusion model (LDM). By capturing patterns and nuances underlying the interdependence of impression opportunities and the multi-agent dynamics of the auction environment, the graph representation enable expressive computations regarding auto-bidding outcomes. With reward alignment techniques, the LDM's posterior is fine-tuned to generate auto-bidding trajectories that maximize KPI metrics while satisfying constraint thresholds. Empirical evaluations on both real-world and synthetic auction environments demonstrate significant improvements in auto-bidding performance across multiple common KPI metrics, as well as accuracy in forecasting auction outcomes.

Problem

• Optimize bidding strategies in competitive auction environments.
• Model large-scale auctions using graph-based diffusion models.
• Maximize KPI metrics while satisfying constraint thresholds.

Innovation

• Graph-based embeddings model auction dynamics
• Latent diffusion model optimizes bidding strategies
• Reward alignment enhances KPI metric performance