Research map / Learning theory and algorithms / Submodular and game algorithms
On Designing the Optimal Integrated Ad Auction in E-commerce Platforms
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 JINTER Net, a RegretNet-style neural auction that jointly allocates ads and organic items in one mixed list to balance revenue and GMV while approximating DSIC and IR.
Problem
Integrated ad auction design for e-commerce search pages, where ads and organic items share slots; ranking them separately is not globally optimal, and classical mechanisms do not account for organic-item externalities while balancing platform revenue and user experience (GMV).
Approach
Treats ads and organic items as bundles of bid and volume nodes in a joint relationship graph, computes allocation probabilities over the combined candidate set via row/column softmax, and sets ad payments via a sigmoid-scaled expected value, trained with an augmented Lagrangian to enforce approximate DSIC and IR.
Key results
- In setting A (2x3x2), JINTER Net reaches Rev+αGMV 1.008 vs 0.971 for IAS and 0.858 for GSP with fixed positions, with regret below 0.001.
- On the Avito dataset, JINTER Net achieves Rev+αGMV 0.948 vs 0.916 for IAS and 0.873 for GSP, a statistically significant gain at p<0.05.
- JINTER Net's Pareto curve dominates GSP and IAS across alpha in [0.5, 2.0] in setting A.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use JINTER Net instead of IAS | Integrated ad auction · Synthetic A: 2 ads, 3 organic items, 2 slots | Rev+αGMV: 0.971 -> 1.008 | Helped |
| Use JINTER Net instead of IAS | Integrated ad auction · Synthetic B: 3 ads, 2 organic items, 2 slots | Rev+αGMV: 0.855 -> 0.939 | Helped |
| Use JINTER Net instead of IAS | Integrated ad auction · Uniform values: 3 ads, 2 organic items, 2 slots | Rev+αGMV: 0.855 -> 0.936 | Helped |
| Use JINTER Net instead of IAS | Integrated ad auction · Normal values: 3 ads, 2 organic items, 2 slots | Rev+αGMV: 0.919 -> 0.925 | Helped |
| Use JINTER Net instead of IAS | Integrated ad auction · Lognormal values: 3 ads, 2 organic items, 2 slots | Rev+αGMV: 0.876 -> 0.920 | Helped |
| Use JINTER Net instead of IAS | Integrated ad auction · Avito test set | Rev+αGMV: 0.916 -> 0.948 | Helped |
Surprising findings
- IAS achieves higher raw revenue than JINTER Net on the Avito dataset (0.308 vs 0.287), so JINTER Net's gain comes from GMV rather than revenue alone.
Limitations
- Evaluated only on small synthetic settings (2-3 ads, 2-3 organic items, 2 slots) and one public dataset with reconstructed CTRs.
- Relies on approximate DSIC via regret thresholds rather than exact truthfulness.
- Volume is treated as a public, fixed quantity and the paper does not study user behavior beyond expected GMV per click.
Experimental setup
Datasets: synthetic value profiles (uniform, truncated normal, truncated lognormal) Avito public dataset
Benchmarks: synthetic auction settings A and B Avito dataset test split
Metrics: social welfare revenue GMV Rev + alpha*GMV ex-post regret
Baselines: GSP with fixed positions IAS
Builds on: RegretNet Myerson auction GSP IAS
Closest papers by meaning
- Merging Mechanisms for Ads and Organic Items in E-commerce Platforms (AAAI 2025 · similarity 0.89)
- Optimal Auction Design in the Joint Advertising (ICML 2025 · similarity 0.87)
- Simultaneous Optimization of Bid Shading and Internal Auction for Demand-Side Platforms (AAAI 2024 · similarity 0.85)
- AuctionNet: A Novel Benchmark for Decision-Making in Large-Scale Games (NeurIPS 2024 · similarity 0.85)
- Maximizing the Success Probability of Policy Allocations in Online Systems (AAAI 2024 · similarity 0.84)
- LLM-Auction: Generative Auction towards LLM-Native Advertising (NeurIPS 2026 · similarity 0.83)
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.