Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contextual Generative Auction with Permutation-level Externalities for Online Advertising

Ruitao Zhu, Yangsu Liu, Dagui Chen, Zhenjia Ma, Chufeng Shi, Zhenzhe Zheng, Jie Zhang, Jian Xu, Bo Zheng, Fan Wu

TL;DR

Contextual Generative Auction is introduced, a novel framework that incorporates permutation-level externalities in multi-slot ad auctions and effectively approximates the optimal auction with nearly maximal revenue and minimal regret.

Abstract

Online advertising has become a core revenue driver for the internet industry, with ad auctions playing a crucial role in ensuring platform revenue and advertiser incentives. Traditional auction mechanisms, like GSP, rely on the independent CTR assumption and fail to account for the influence of other displayed items, termed externalities. Recent advancements in learning-based auctions have enhanced the encoding of high-dimensional contextual features. However, existing methods are constrained by the "allocation-after-prediction" design paradigm, which models set-level externalities within candidate ads and fails to consider the sequential context of the final allocation, leading to suboptimal results. This paper introduces the Contextual Generative Auction (CGA), a novel framework that incorporates permutation-level externalities in multi-slot ad auctions. Built on the structure of our theoretically derived optimal solution, CGA decouples the optimization of allocation and payment. We construct an autoregressive generative model for allocation and reformulate the incentive compatibility (IC) constraint into minimizing ex-post regret that supports gradient computation, enabling end-to-end learning of the optimal payment rule. Extensive offline and online experiments demonstrate that CGA significantly enhances platform revenue and CTR compared to existing methods, while effectively approximating the optimal auction with nearly maximal revenue and minimal regret.

Contextual Generative Auction with Permutation-level Externalities for Online Advertising

TL;DR

Contextual Generative Auction is introduced, a novel framework that incorporates permutation-level externalities in multi-slot ad auctions and effectively approximates the optimal auction with nearly maximal revenue and minimal regret.

Abstract

Online advertising has become a core revenue driver for the internet industry, with ad auctions playing a crucial role in ensuring platform revenue and advertiser incentives. Traditional auction mechanisms, like GSP, rely on the independent CTR assumption and fail to account for the influence of other displayed items, termed externalities. Recent advancements in learning-based auctions have enhanced the encoding of high-dimensional contextual features. However, existing methods are constrained by the "allocation-after-prediction" design paradigm, which models set-level externalities within candidate ads and fails to consider the sequential context of the final allocation, leading to suboptimal results. This paper introduces the Contextual Generative Auction (CGA), a novel framework that incorporates permutation-level externalities in multi-slot ad auctions. Built on the structure of our theoretically derived optimal solution, CGA decouples the optimization of allocation and payment. We construct an autoregressive generative model for allocation and reformulate the incentive compatibility (IC) constraint into minimizing ex-post regret that supports gradient computation, enabling end-to-end learning of the optimal payment rule. Extensive offline and online experiments demonstrate that CGA significantly enhances platform revenue and CTR compared to existing methods, while effectively approximating the optimal auction with nearly maximal revenue and minimal regret.

Paper Structure

This paper contains 25 sections, 4 theorems, 36 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. PRELIMINARIES
  3. Multi-slot Auctions with Permutation-level Externalities
  4. Optimal Multi-slot Auction
  5. Auction Design as a Learning-based Problem
  6. Methodology
  7. Generator: Autoregressive Generative Module
  8. Evaluator: Permutation-aware Prediction Module
  9. Payment Computation
  10. Optimization and Training
  11. Optimization of G-E Framework
  12. Optimization of PaymentNet
  13. Experiments
  14. Experiment Setup
  15. Dataset
  16. ...and 10 more sections

Key Result

Theorem 1

(Myerson's Lemma myerson1981optimal). For a single-parameter environment, an allocation rule $\mathcal{A}$ is implementable if there exists a payment rule $\mathcal{P}$ such the mechanism $\mathcal{M}\langle \mathcal{A},\mathcal{P}\rangle$ is DSIC. The following two claims hold: (1) An allocation ru

Figures (2)

  • Figure 1: The Permutation-aware ad CTR as a function of slot on Taobao is non-monotonic.
  • Figure 2: The architecture of Contextual Generative Auction (CGA). The middle part shows the overall framework of CGA, with dashed outlines and arrows depicting offline training components, and solid lines representing online inference paths. The other two parts provide specific implementations of Generator and Evaluator.

Theorems & Definitions (7)

  • Definition 1
  • Definition 2
  • Definition 3
  • Theorem 1
  • Lemma 1
  • Theorem 2
  • Corollary 1