ACQ: A Unified Framework for Automated Programmatic Creativity in Online Advertising

TL;DR

The paper addresses maximizing DSP revenue by optimally allocating the number of creatives per photo, a problem with finite quotas and diminishing returns. It introduces ACQ, a two-stage framework consisting of a prediction module based on an Unbalanced Binary Tree Model (UBTM) with multi-task learning and network-property constraints, and an allocation module solving a large-scale multiple-choice knapsack problem via a Lagrangian dual approach with a dual-based binary search solver (DBSSolver). The approach enforces monotonicity, submodularity, and smoothness to improve interpretability and stability, and demonstrates scalable performance on tens of millions of ads. Offline and online experiments on Kuaishou show meaningful gains, including a 9.34% increase in cost and a 1.42% rise in creatives, highlighting practical value for automated creative management in programmatic advertising.

Abstract

In online advertising, the demand-side platform (a.k.a. DSP) enables advertisers to create different ad creatives for real-time bidding. Intuitively, advertisers tend to create more ad creatives for a single photo to increase the probability of participating in bidding, further enhancing their ad cost. From the perspective of DSP, the following are two overlooked issues. On the one hand, the number of ad creatives cannot grow indefinitely. On the other hand, the marginal effects of ad cost diminish as the number of ad creatives increases. To this end, this paper proposes a two-stage framework named Automated Creatives Quota (ACQ) to achieve the automatic creation and deactivation of ad creatives. ACQ dynamically allocates the creative quota across multiple advertisers to maximize the revenue of the ad platform. ACQ comprises two components: a prediction module to estimate the cost of a photo under different numbers of ad creatives, and an allocation module to decide the quota for photos considering their estimated costs in the prediction module. Specifically, in the prediction module, we develop a multi-task learning model based on an unbalanced binary tree to effectively mitigate the target variable imbalance problem. In the allocation module, we formulate the quota allocation problem as a multiple-choice knapsack problem (MCKP) and develop an efficient solver to solve such large-scale problems involving tens of millions of ads. We performed extensive offline and online experiments to validate the superiority of our proposed framework, which increased cost by 9.34%.

Figures (5)

• Figure 1: The hierarchical relationships within advertising. From top to bottom, these levels are campaigns, units, and creatives. A creative can be created from only one photo. However, advertisers can delegate this process to the advertising platform, which can automatically create multiple creatives from a single photo by combining different titles or themes.
• Figure 2: Binary tree modeling on real-world datasets. The leaf nodes may prevent the entire structure from being a strict binary tree due to the significant long-tail distribution and the overlap of quantiles, which result in indivisible sub-intervals.
• Figure 3: Our prediction model structure includes essential components and extensible property exploration components.
• Figure 4: The actual online operation framework.
• Figure 5: Parameter sensitivity experiments.