Research map / Probabilistic and scientific modeling / Molecules and proteins
GenAuction: A Generative Auction for Online Advertising
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes GenAuction, a KPI-based, PV-level generative auction that uses a Transformer Generator-Evaluator with VCG payments and DDPG-style training, evaluated on Baidu auction logs and an online A/B test.
Problem
Rule-based multi-objective ad auctions select ads individually with fixed rank-score aggregation, which is computationally heavy and suboptimal across multi-round auctions and ignores within-page-view (PV) contextual effects between ads.
Approach
Formulate ad allocation as a constrained optimization over KPIs and select winning page views (PVs) directly at the PV level using a Transformer Generator that proposes candidate PVs and an Evaluator that picks the best one with full-PV context, paid via VCG.
Key results
- Online A/B test over 30 days with 10% traffic shows +8.1% social welfare versus the tVCG baseline while constraints are maintained.
- Offline, GenAuction's SW-versus-constraint curves lie above uGSP, tVCG and EdgeNet and just below the VCG upper bound.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use GenAuction instead of tVCG | Baidu ad auction · Baidu online A/B test | social welfare: +8.1% | Helped |
Limitations
- VCG is only a theoretical optimum because enumerating candidate PV combinations is computationally infeasible at industrial scale.
- Evaluation relies on a proprietary Baidu dataset and a single online A/B test with one baseline (tVCG).
- Payment focuses on VCG and the paper does not thoroughly analyze the alternative payment schemes it mentions.
Experimental setup
Datasets: Baidu 30-day online auction log (500 advertisers, 5.5M auction rounds)
Benchmarks: Baidu auction log offline evaluation online A/B test on Baidu
Metrics: social welfare PV relevance show ratio ad show ratio
Baselines: uGSP VCG tVCG EdgeNet
Builds on: VCG mechanism GSP EdgeNet DDPG position-aware context aggregation
Closest papers by meaning
- LLM-Auction: Generative Auction towards LLM-Native Advertising (NeurIPS 2026 · similarity 0.90)
- AuctionNet: A Novel Benchmark for Decision-Making in Large-Scale Games (NeurIPS 2024 · similarity 0.87)
- Ad Auctions for LLMs via Retrieval Augmented Generation (NeurIPS 2024 · similarity 0.87)
- Maximizing the Success Probability of Policy Allocations in Online Systems (AAAI 2024 · similarity 0.85)
- Truthful Aggregation of LLMs with an Application to Online Advertising (NeurIPS 2025 · similarity 0.84)
- Learning Personalized Ad Impact via Contextual Reinforcement Learning under Delayed Rewards (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.