Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BAT: Benchmark for Auto-bidding Task

Alexandra Khirianova, Ekaterina Solodneva, Andrey Pudovikov, Sergey Osokin, Egor Samosvat, Yuriy Dorn, Alexander Ledovsky, Yana Zenkova

TL;DR

BAT addresses the scarcity of reliable autobidding benchmarks in real-time bidding by providing a large-scale, two-auction RTB dataset (FP and VCG) from Avito, along with three data components (campaigns, auction statistics, traffic) and comprehensive quality filtering. It introduces a set of baselines—ALM, TA-PID, M-PID, Mystique, and BROI—plus improvements to M-PID, and evaluates them through three experimental lenses: budget pacing, CPC constraints, and click maximization. The work demonstrates how traffic patterns and auction formats influence bidding performance and offers reproducible code and evaluation frameworks to foster further innovation in autobidding methods. Overall, BAT serves as a practical, open benchmark that connects real-world platform dynamics with controllable simulations to accelerate development in programmatic advertising research.

Abstract

The optimization of bidding strategies for online advertising slot auctions presents a critical challenge across numerous digital marketplaces. A significant obstacle to the development, evaluation, and refinement of real-time autobidding algorithms is the scarcity of comprehensive datasets and standardized benchmarks. To address this deficiency, we present an auction benchmark encompassing the two most prevalent auction formats. We implement a series of robust baselines on a novel dataset, addressing the most salient Real-Time Bidding (RTB) problem domains: budget pacing uniformity and Cost Per Click (CPC) constraint optimization. This benchmark provides a user-friendly and intuitive framework for researchers and practitioners to develop and refine innovative autobidding algorithms, thereby facilitating advancements in the field of programmatic advertising. The implementation and additional resources can be accessed at the following repository (https://github.com/avito-tech/bat-autobidding-benchmark, https://doi.org/10.5281/zenodo.14794182).

BAT: Benchmark for Auto-bidding Task

TL;DR

BAT addresses the scarcity of reliable autobidding benchmarks in real-time bidding by providing a large-scale, two-auction RTB dataset (FP and VCG) from Avito, along with three data components (campaigns, auction statistics, traffic) and comprehensive quality filtering. It introduces a set of baselines—ALM, TA-PID, M-PID, Mystique, and BROI—plus improvements to M-PID, and evaluates them through three experimental lenses: budget pacing, CPC constraints, and click maximization. The work demonstrates how traffic patterns and auction formats influence bidding performance and offers reproducible code and evaluation frameworks to foster further innovation in autobidding methods. Overall, BAT serves as a practical, open benchmark that connects real-world platform dynamics with controllable simulations to accelerate development in programmatic advertising research.

Abstract

The optimization of bidding strategies for online advertising slot auctions presents a critical challenge across numerous digital marketplaces. A significant obstacle to the development, evaluation, and refinement of real-time autobidding algorithms is the scarcity of comprehensive datasets and standardized benchmarks. To address this deficiency, we present an auction benchmark encompassing the two most prevalent auction formats. We implement a series of robust baselines on a novel dataset, addressing the most salient Real-Time Bidding (RTB) problem domains: budget pacing uniformity and Cost Per Click (CPC) constraint optimization. This benchmark provides a user-friendly and intuitive framework for researchers and practitioners to develop and refine innovative autobidding algorithms, thereby facilitating advancements in the field of programmatic advertising. The implementation and additional resources can be accessed at the following repository (https://github.com/avito-tech/bat-autobidding-benchmark, https://doi.org/10.5281/zenodo.14794182).
Paper Structure (26 sections, 11 equations, 5 figures, 10 tables, 2 algorithms)

This paper contains 26 sections, 11 equations, 5 figures, 10 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Contribution
  3. Related works
  4. Dataset description
  5. Campaigns
  6. Auction statistics
  7. Traffic
  8. Data collection, filtering and sampling
  9. Dataset statistics
  10. Problem formulation
  11. Baselines
  12. Adaptive Linear Model (ALM)
  13. Traffic-aware PID (TA-PID)
  14. Model predictive PID (M-PID)
  15. Mystique
  16. ...and 11 more sections

Figures (5)

  • Figure 1: The statistics on week traffic distribution for all regions on average.
  • Figure 2: Dependences of AuctionClicksSurplus and AuctionWinBidSurplus on contact price bin for an example of micricategory.
  • Figure 3: Dependence of contact price bin on hour: Blue shades represent VCG, while orange represents FP. The graph shows the time-averaged price bin, standard deviation, and one example of values for a specific campaign for each auction type.
  • Figure 4: The number of campaigns per logical category.
  • Figure 5: Traffic definitions.