Reproduce “MEBS: Multi-task End-to-end Bid Shading for Multi-slot Display Advertising” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhen Gong, Lvyin Niu, Yang Zhao, Miao Xu, Zhenzhe Zheng, Haoqi Zhang, Zhilin Zhang, Fan Wu and 4 more
The MEBS model optimizes bid shading for multi-slot display advertising by formulating an end-to-end multi-task learning approach to maximize advertiser expected surplus. It achieved a 7.01% increase in Gross Merchandise Volume and a 7.42% boost in Return on Investment in online A/B tests on a large-scale platform.
Key insights
- The optimal bidding strategy in multi-slot display advertising can be theoretically decoupled into budget pacing and per-impression surplus maximization, where surplus depends on the bid's impact on winning probability and position-calibrated CTR.
- Multi-task learning with shared feature embeddings effectively addresses data sparsity in modeling bid-aware win rates and calibrated CTRs, which are crucial components for accurate bid shading.
- Directly optimizing for the expected advertiser surplus in an end-to-end manner provides a robust solution when optimal bid price labels are not explicitly available, leading to superior performance compared to methods relying on surrogate losses.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.