EVENT DOSSIER
On September 7, 2026, NetEase Game Community Application DASHEN launched the system called AutoLR, designed to automate the entire process of industrial recommendation systems from research to launch review. This system integrates three mechanisms: debate and review by multiple expert committees, weighted exploration using a selector based on definitive evidence, and fusion of external research and in-domain data through a hierarchical knowledge system. The LLM agent is responsible for semantic reasoning and code generation, while the definitive controller retains authority over execution, metric extraction, and state transition. AutoLR transforms the long-term experimental processes that previously relied on manual coordination into autonomous closed loops, covering the entire automation of processes from paper reproduction, offline evaluation, to A/B testing and launch review.
网易游戏社区应用 DASHEN 推出 AutoLR，旨在自动化从研究到上线评审的工业推荐系统流程。该系统整合三大机制：多专家委员会辩论评审、确定性证据加权探索 - 利用选择器分配试验预算及分层知识系统融合外部研究与域内数据。LLM 代理负责语义推理与代码生成，而确定性控制器保留执行、指标提取及状态转换的权威。AutoLR 将原本依赖人工协调的长周期实验流程转化为自主闭环，涵盖从论文复现、离线评估到 A/B 测试及上线评审的全链路自动化。