Reproduce “MTMD: A Multi-Task Multi-Domain Framework for Unified Ad Lightweight Ranking at Pinterest” 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
Xiao Yang, Peifeng Yin, Abe Engle, Jinfeng Zhuang, Ling Leng
Pinterest researchers developed MTMD, a unified multi-task multi-domain framework for lightweight ad ranking, which successfully replaced nine separate production models. This framework improved key online metrics, including a 1.96% reduction in Cost Per Click and a 2.41% increase in Click-Through Rate.
Key insights
- Consolidating disparate ad ranking models into a single framework, empowered by sophisticated multi-task multi-domain learning, yields superior performance over fragmented approaches.
- Mixture-of-Experts architectures can effectively balance the learning of specialized knowledge for individual tasks/domains with generalized knowledge across the system.
- Designing for explicit task dependencies and aligning lightweight ranking with downstream heavyweight scores through label choice are crucial for optimizing cascaded ranking systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.