Deep Automated Mechanism Design for Integrating Ad Auction and Allocation in Feed

Xuejian Li,Ze Wang,Bingqi Zhu,Fei He,Yongkang Wang,Xingxing Wang

2024-04-11

Abstract:E-commerce platforms usually present an ordered list, mixed with several organic items and an advertisement, in response to each user's page view request. This list, the outcome of ad auction and allocation processes, directly impacts the platform's ad revenue and gross merchandise volume (GMV). Specifically, the ad auction determines which ad is displayed and the corresponding payment, while the ad allocation decides the display positions of the advertisement and organic items. The prevalent methods of segregating the ad auction and allocation into two distinct stages face two problems: 1) Ad auction does not consider externalities, such as the influence of actual display position and context on ad Click-Through Rate (CTR); 2) The ad allocation, which utilizes the auction-winning ad's payment to determine the display position dynamically, fails to maintain incentive compatibility (IC) for the advertisement. For instance, in the auction stage employing the traditional Generalized Second Price (GSP) , even if the winning ad increases its bid, its payment remains unchanged. This implies that the advertisement cannot secure a better position and thus loses the opportunity to achieve higher utility in the subsequent ad allocation stage. Previous research often focused on one of the two stages, neglecting the two-stage problem, which may result in suboptimal outcomes...

Computer Science and Game Theory,Artificial Intelligence

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

This paper aims to address two main issues in the ad auction and allocation process on e-commerce platforms:

1. **Ad auctions do not consider externalities**: Traditional ad auction methods (such as Generalized Second Price Auction, GSP) usually do not take into account the impact of ad display position and context on click-through rate (CTR). This leads to complex strategic competition, failing to achieve stable outcomes and maximization of social welfare.

2. **Dynamic ad allocation is not incentive-compatible**: In the ad auction phase, even if the winning ad increases its bid, its payment amount does not change, thus it cannot obtain a better display position, losing the opportunity to gain higher utility in the subsequent ad allocation phase. Moreover, using the winning ad's bid to determine the display position is also inappropriate, as it may lead to a loss in the platform's total transaction volume (GMV) exceeding the increase in revenue.

To address these issues, the paper proposes a deep automated mechanism design method (MIAA), which integrates ad auction and allocation to ensure incentive compatibility and individual rationality in the presence of externalities, while maximizing the platform's revenue and GMV. Specifically, MIAA achieves this goal through the following steps:

- **Generate candidate allocations**: For each candidate ad, generate multiple candidate allocations by inserting it into different positions in the organic item list.
- **Model global externalities**: Use a list-level prediction model to input each candidate allocation and output the predicted results for each ad and organic item to model global externalities.
- **Execute automated auction mechanism**: Select the optimal allocation through an automated auction mechanism modeled by deep neural networks, while determining the ranking, payment, and display position of the ads.

Through this method, MIAA not only addresses the issues present in traditional methods but also demonstrates higher revenue and GMV compared to existing baseline methods in offline experiments and online A/B tests.