Reproduce “AgenticRecTune: Multi-Agent with Self-Evolving Skillhub for Recommendation System Optimization” 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
Xidong Wu, Yue Zhuan, Ruoqiao Wei, Hangxin Chen, Di Bai, Jintao Liu, Xinyi Wang, Xue Wang and 2 more
Google's AgenticRecTune is a multi-agent framework designed to automate and optimize system-level configurations within large-scale, multi-stage recommender systems. It leverages a self-evolving knowledge base and iterative A/B testing, achieving statistically significant online improvements in Google Discover, such as a 0.75% increase in Engagement Metric 1 for pre-ranking and a 3.43% increase in Diversity for re-ranking.
Key insights
- A multi-agent architecture (Actor-Critic) significantly enhances optimization precision and robustness by iteratively refining proposals, outperforming single-agent approaches for complex, non-differentiable systems.
- Integrating a self-evolving knowledge base (Skillhub) allows the system to autonomously learn, refine domain expertise, and adapt to evolving business objectives without manual intervention.
- Direct optimization through online A/B testing, orchestrated by an Online Agent, effectively bridges the gap between offline proxy metrics and real-world 'north star' business performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.