跳转至

文章背景与核心概要

在大规模数字广告系统中，需求方平台（DSP）允许广告主上传单张图片并生成多个广告创意，以提升实时竞价（RTB）的机会。然而，无限增加创意数量会导致边际效益递减，因此无法实现无限扩展。为了解决这一痛点，快手研究团队提出了 ACQ（Automated Creative Quota，自动化创意配额） 框架。

该文提出了一种工业界落地的两阶段解决方案：第一阶段利用基于非平衡二叉树的多任务模型来预测配额条件下的期望收益，专门用于应对跨配额级别的极度偏斜收益分布；第二阶段将带有全局容量约束的配额分配问题建模为多重背包问题（MCKP），并通过基于对偶理论的算法进行求解。在快手广告投放平台上的大规模线上线下实验表明，ACQ 框架成功实现了平台广告收入 6.20% 的显著增长。

执行摘要 (Executive Summary)

This paper presents ACQ (Automated Creative Quota), a deployed two-stage framework designed to solve the challenge of creative quota allocation in large-scale online advertising. In digital advertising, demand-side platforms (DSPs) allow advertisers to upload photos and generate multiple ad creatives to improve real-time bidding opportunities. However, generating more creatives exhibits diminishing returns and cannot scale infinitely. ACQ addresses this by predicting quota-conditioned expected revenue using an unbalanced binary tree-based multi-task model, and subsequently formulating global capacity constraints as a multiple-choice knapsack problem (MCKP) solved via a dual-based algorithm. Deployed on Kuaishou's advertising platform, ACQ has successfully achieved a 6.20% increase in platform advertising revenue.

本文提出了 ACQ（自动化创意配额），这是一个已实际部署的两阶段框架，旨在解决大规模在线广告中创意配额分配的挑战。在数字广告中，需求方平台（DSP）允许广告主上传图片并生成多个广告创意，以提高实时竞价的机会。然而，生成更多的创意会呈现边际效益递减，无法无限扩展。ACQ 通过使用基于非平衡二叉树的多任务模型来预测配额条件下的期望收益，并随后将全局容量约束建模为多重选择背包问题（MCKP）并通过基于对偶算法进行求解，从而解决了这一问题。ACQ 已部署在快手的广告平台上，成功实现了平台广告收入 6.20% 的增长。

论文概览 (Paper Overview)

基本信息 (Bibliographic Information)

  • Title: ACQ: A Deployed Two-Stage Framework for Automated Creative Quota Allocation in Large-Scale Online Advertising
  • Authors: Ruizhi Wang, Yu Rong, Kai Liu, Bingjie Li, Qingpeng Cai, Fei Pan, Peng Jiang
  • Primary Subject: Artificial Intelligence (cs.AI)
  • arXiv Identifier: arXiv:2412.06167
  • Submitted On: December 9, 2024 (Last revised August 21, 2026)
  • 标题： ACQ：大规模在线广告中自动化创意配额分配的落地两阶段框架
  • 作者： Ruizhi Wang, Yu Rong, Kai Liu, Bingjie Li, Qingpeng Cai, Fei Pan, Peng Jiang
  • 主要学科： 人工智能 (cs.AI)
  • arXiv 标识符： arXiv:2412.06167
  • 提交时间： 2024年12月9日（最近修订于2026年8月21日）

摘要 (Abstract)

In digital advertising, demand-side platforms (DSPs) allow advertisers to create multiple ad creatives from a single photo for real-time bidding. While increasing the number of creatives can improve bidding opportunities, it cannot scale indefinitely, and the incremental advertising revenue typically exhibits diminishing returns as more creatives are generated.

在数字广告中，需求方平台（DSP）允许广告主从单张图片中创建多个广告创意以进行实时竞价。虽然增加创意数量可以改善竞价机会，但它无法无限扩展，并且随着生成更多创意，增量广告收入通常会表现出边际效益递减。

This raises a practical problem for DSPs: how to automatically determine an appropriate creative quota for each photo at scale. To address this problem, we propose Automated Creatives Quota (ACQ), a deployed two-stage framework for creative quota allocation in large-scale online advertising.

这给 DSP 带来了一个实际问题：如何在规模化场景下自动为每张图片确定合适的创意配额。为了解决这个问题，我们提出了 自动化创意配额（ACQ），这是一个用于大规模在线广告中创意配额分配的落地两阶段框架。

  • Stage 1 (Prediction): ACQ predicts quota-conditioned expected revenue using a multi-task model built on an unbalanced binary tree, designed specifically to handle the highly skewed revenue distribution across quota levels.
  • Stage 2 (Optimization): ACQ formulates quota allocation under global capacity constraints as a multiple-choice knapsack problem (MCKP) and solves it efficiently using a dual-based algorithm.
  • 第一阶段（预测）： ACQ 使用构建在非平衡二叉树之上的多任务模型来预测配额条件下的期望收益，该模型专门用于处理跨配额级别高度偏斜的收益分布。
  • 第二阶段（优化）： ACQ 将全局容量约束下的配额分配建模为多重选择背包问题（MCKP），并使用基于对偶的算法高效地求解。

核心结果与性能 (Key Results & Performance)

  • Platform Impact: Extensive offline and online experiments conducted on Kuaishou's advertising delivery platform demonstrate the real-world efficacy of the framework.
  • Revenue Growth: The deployed ACQ framework achieved a notable 6.20% increase in platform advertising revenue.
  • 平台影响： 在快手广告投放平台上进行的广泛离线和在线实验证明了该框架的实际有效性。
  • 收入增长： 已部署的 ACQ 框架实现了平台广告收入显著增长 6.20%。