论文

面向大规模排序模型特征选优的 Agentic 系统在 arXiv 公开

Cost-Aware Agentic Feature Optimization System for Fleet-Scale Ranking Models

精选理由

大平台用 LLM agent 给成排排序模型自动挑特征,交叉特征生成加 10–25 轮异步调优,normalized entropy 平均改善 0.31%,流水线成本还降了 2.3%。做推荐系统或特征工程的人可以看看这套 LLM 提假设、工具做验证的分工方式。

一篇 arXiv 论文提出端到端 agentic 系统,用于优化大型 Web 平台成排排序模型的特征选择,这些模型从数万个共享特征池中选取特征,需同时满足跨模型共享的和单模型各自的资源约束,如特征流水线成本和服务吞吐 QPS。系统由 LLM 智能体提出假设、确定性工具执行实验并验证:LLM 从组合候选空间生成并初筛交叉特征,多种重要性方法对所有特征排序,特征融合混合各排序结果,多任务模型用跨任务的 Pareto 前沿处理。随后 LLM Researcher agent 在资源上限内用 10–25 个异步试验调优特征选择,裁剪步骤把流水线成本压回预算内。

正文

在生产模型上,系统在各模型预算内平均将 normalized entropy 改善 0.31%,裁剪步骤使流水线成本降低 2.3%,且回归幅度不超出测量噪声。优化后的模型已上线,在线 A/B 测试中主要线上指标提升 0.20%。论文展示的是 LLM 假设生成与确定性实验执行相结合的工程范式,对维护大规模特征工程流水线的推荐系统团队有直接参考价值。

原文 · arXiv cs.AI

Cost-Aware Agentic Feature Optimization System for Fleet-Scale Ranking Models

Large Web platforms operate large fleets of ranking models, each selecting features from a shared pool of tens of thousands. We present an end-to-end agentic system that optimizes features across the fleet under resource constraints shared across models and those set per model, such as feature pipeline cost and serving throughput (QPS), respectively. In this system, LLM agents propose hypotheses, and deterministic tools execute experiments and verify them. For each model, an LLM generates and shortlists cross features from a combinatorial candidate space, and all features are then ranked by several importance methods. Feature blending mixes these rankings into a single selection, using Pareto fronts across tasks for multi-task models. An LLM Researcher agent then tunes the feature selection in 10--25 asynchronous trials under resource caps, and a trimming step keeps the pipeline cost within budget. Across production models, the system improves normalized entropy by 0.31% on average within each model's budget, and the trimming step cuts pipeline cost by 2.3% with no regression beyond measurement noise. The optimized models were launched, and in online A/B tests they improved a primary online metric by 0.20%.

产业坐标与信源核验

本条情报归属 AITOP「论文」核心栏目,重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:Agentic System、推荐系统、特征工程、排序模型,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始线索采编自「arXiv cs.AI」(@Chang Yang, Yao Xuan, Wendan Yan, Nhat Le, Jiang Liu, Longhao Jin, Zhen Peggy Yao, Alireza Vahdatpour, Djordje Gligorijevic),经 AITOP 多维评分管线结构化提炼。