Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Joint Auction in the Online Advertising Market

Zhen Zhang, Weian Li, Yahui Lei, Bingzhe Wang, Zhicheng Zhang, Qi Qi, Qiang Liu, Xingxing Wang

TL;DR

This work tackles revenue optimization in online advertising when stores and brand suppliers bid jointly for ad slots, formulating a DSIC and IR-constrained mechanism design and recasting it as a learning problem. It introduces JRegNet, a neural network with an allocation component that produces a bi-stochastic bundle allocation $A$ and a payment component that yields $P$, all guided by a joint relationship graph and constrained via an augmented Lagrangian and misreport optimization. The model explicitly handles bundle formation, correlated bids, and per-bidder payments, enabling near-optimal revenue through automated mechanism design. Empirical results on synthetic data and real Meituan logs show that JRegNet achieves higher revenue than RegretNet, IRegNet, and VCG while maintaining low ex-post regret, demonstrating practical viability for joint advertising ecosystems.

Abstract

Online advertising is a primary source of income for e-commerce platforms. In the current advertising pattern, the oriented targets are the online store owners who are willing to pay extra fees to enhance the position of their stores. On the other hand, brand suppliers are also desirable to advertise their products in stores to boost brand sales. However, the currently used advertising mode cannot satisfy the demand of both stores and brand suppliers simultaneously. To address this, we innovatively propose a joint advertising model termed Joint Auction, allowing brand suppliers and stores to collaboratively bid for advertising slots, catering to both their needs. However, conventional advertising auction mechanisms are not suitable for this novel scenario. In this paper, we propose JRegNet, a neural network architecture for the optimal joint auction design, to generate mechanisms that can achieve the optimal revenue and guarantee near dominant strategy incentive compatibility and individual rationality. Finally, multiple experiments are conducted on synthetic and real data to demonstrate that our proposed joint auction significantly improves platform revenue compared to the known baselines.

Joint Auction in the Online Advertising Market

TL;DR

This work tackles revenue optimization in online advertising when stores and brand suppliers bid jointly for ad slots, formulating a DSIC and IR-constrained mechanism design and recasting it as a learning problem. It introduces JRegNet, a neural network with an allocation component that produces a bi-stochastic bundle allocation and a payment component that yields , all guided by a joint relationship graph and constrained via an augmented Lagrangian and misreport optimization. The model explicitly handles bundle formation, correlated bids, and per-bidder payments, enabling near-optimal revenue through automated mechanism design. Empirical results on synthetic data and real Meituan logs show that JRegNet achieves higher revenue than RegretNet, IRegNet, and VCG while maintaining low ex-post regret, demonstrating practical viability for joint advertising ecosystems.

Abstract

Online advertising is a primary source of income for e-commerce platforms. In the current advertising pattern, the oriented targets are the online store owners who are willing to pay extra fees to enhance the position of their stores. On the other hand, brand suppliers are also desirable to advertise their products in stores to boost brand sales. However, the currently used advertising mode cannot satisfy the demand of both stores and brand suppliers simultaneously. To address this, we innovatively propose a joint advertising model termed Joint Auction, allowing brand suppliers and stores to collaboratively bid for advertising slots, catering to both their needs. However, conventional advertising auction mechanisms are not suitable for this novel scenario. In this paper, we propose JRegNet, a neural network architecture for the optimal joint auction design, to generate mechanisms that can achieve the optimal revenue and guarantee near dominant strategy incentive compatibility and individual rationality. Finally, multiple experiments are conducted on synthetic and real data to demonstrate that our proposed joint auction significantly improves platform revenue compared to the known baselines.
Paper Structure (16 sections, 1 theorem, 18 equations, 6 figures, 5 tables, 1 algorithm)

This paper contains 16 sections, 1 theorem, 18 equations, 6 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Main Contributions
  3. Related Work
  4. Joint Auction Design
  5. Joint Auction
  6. Joint Auction Design as a Learning Problem
  7. JRegNet
  8. The JRegNet Architecture
  9. Training of JRegNet
  10. Experiments
  11. Synthetic Data
  12. Real-World Dataset
  13. Conclusions
  14. Additional Experimental Information
  15. Detailed Experiment Results for Different CTRs and Different Numbers of Advertisement Slots.
  16. ...and 1 more sections

Key Result

Lemma 1

The matrix $\varphi_{q k}^{B S}\left(c, r\right)$ is bi-stochastic $\forall c, r \in \mathbb{R}^{Q K}$. For any bi-stochastic matrix $a \in[0,1]^{Q K}$, $\exists c, r \in \mathbb{R}^{Q K}$ for which $a=\varphi_{q k}^{B S}\left(c, r\right)$.

Figures (6)

  • Figure 1: An Example of the Traditional Advertising Model and Joint Advertising Model.
  • Figure 2: The bundle relationship between stores and brands. We use different beverage icons to represent different brands, and an edge linking a store and a brand means that there is a sales relationship between the two.
  • Figure 3: The JRegNet architecture, including the allocation network part and the payment network part, for a setting of $m$ stores, $n$ brands, $K$ advertising slots, and $Q$ bundles of stores and brands.
  • Figure 4: The results of experiments on the settings of different CTRs (a) and different numbers of advertisement slots (b). In all results, the regrets are less than 0.001.
  • Figure 5: The number of real search query samples used for training JRegNet. The horizontal axis is the generation time of the search query samples. In this figure, all search query samples are obtained from logs of the year 2023. Within settings E1 to E3, we use data from (a) to (c) to train JRegNet, respectively.
  • ...and 1 more figures

Theorems & Definitions (3)

  • Definition 1: DSIC
  • Definition 2: IR
  • Lemma 1: dutting2019optimal