← 研踪 PaperTrailarXiv:2611.22503 · v1
From Relevance to Utility: Rubric Alignment in Yuanbao Search
首次提交 2026-11-17 18:00
版本更新 2026-11-17 18:00
雷达观测 2026-11-18 11:33
摘要
Agentic search demands that a search engine move beyond per document relevance ranking to assemble document sets optimized for downstream LLM reasoning. Prior work addresses this either by improving individual pipeline stages in isolation or by applying RL based selection on top of relevance trained upstream modules, leaving the full stack unaligned to utility for reasoning. We propose rubric aligned search, a framework that unifies the traditional search pipeline under a single supervision surface. For each query, we derive a rubric from coarse to fine, specifying weighted positive criteria that the document set must satisfy and negative criteria that it must avoid. From this rubric we derive two training signals. First, a lightweight labeler distills rubric grounded judgments into per document utility labels, which supervise a contrastive dual encoder for retrieval and a multi task ranker trained with four rubric aligned scoring heads. Second, a rubric conditioned evaluator produces a set level reward that drives a dual branch RL selector, combining GRPO with best rollout distillation to resolve credit assignment in subset optimization. Deployed in Yuanbao Search, the pipeline improves end to end answer quality by 8.52% over the online production baseline.
作者备注:10 pages, 5 figures, 5 tables, accepted at KDD 2027
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-17 18:00 · 最新观测版本
From Relevance to Utility: Rubric Alignment in Yuanbao Search
Agentic search demands that a search engine move beyond per document relevance ranking to assemble document sets optimized for downstream LLM reasoning. Prior work addresses this either by improving individual pipeline stages in isolation or by applying RL based selection on top of relevance trained upstream modules, leaving the full stack unaligned to utility for reasoning. We propose rubric aligned search, a framework that unifies the traditional search pipeline under a single supervision surface. For each query, we derive a rubric from coarse to fine, specifying weighted positive criteria that the document set must satisfy and negative criteria that it must avoid. From this rubric we derive two training signals. First, a lightweight labeler distills rubric grounded judgments into per document utility labels, which supervise a contrastive dual encoder for retrieval and a multi task ranker trained with four rubric aligned scoring heads. Second, a rubric conditioned evaluator produces a set level reward that drives a dual branch RL selector, combining GRPO with best rollout distillation to resolve credit assignment in subset optimization. Deployed in Yuanbao Search, the pipeline improves end to end answer quality by 8.52% over the online production baseline.
查看此版本解读 →查看此版本原文 ↗