arXiv cs.CL· Yunhan Li, Mingjie Xie, Zeyang Shi, Gengshen Wu, Min Yang·· 13 小时前AI 评分29
超越召回饱和：面向工业法律检索的难负样本重排序器训练扩展
Beyond Recall Saturation: Scaling Hard-Negative Reranker Training for Industrial Legal Retrieval
AI 导读
在中文法律检索基准上，embedding top-300 已进入高召回区间，但相关证据仍可能排在输出截断线以下，说明瓶颈转向重排序器质量。研究对比 embedding 挖掘、自挖掘与动态负样本等难负样本策略，发现 embedding-to-self-mined 两阶段流程性价比最优，额外动态阶段并未提升全部留出指标。
正文 · 原文
Abstract:Industrial retrieval systems often improve quality by enlarging embedding candidate pools or adopting complex training pipelines. We show that, under a sufficiently large but practical candidate pool, embedding recall exhibits diminishing returns, motivating a closer examination of reranker quality. On Chinese legal retrieval benchmarks, embedding top-300 reaches a high-recall operating region, while relevant evidence may still be ranked below the output cutoff. We compare hard-negative strategies for reranker training, including embedding-mined, self-mined, and dynamic-negative variants, under aligned retrieval protocols that evaluate reranking after candidate retrieval. Among the evaluated staged configurations, an embedding-to-self-mined pipeline provides a favorable trade-off; additional dynamic stages do not improve all held-out metrics. The selected pipeline scales from a 0.6B to a 4B reranker. Relative to the native 4B reranker, the trained 4B Stage2 model improves reported metrics on the STARD and CSAID benchmarks and on an internal development set, and is preferred in 72.4\% of non-tied comparisons on 410 de-identified, reconstructed internal operational-query inputs in a latency-sensitive offline preference evaluation. These results support evaluating the retrieval and reranking stages separately when selecting an optimization target for industrial retrieval systems.
| Comments: | 11 pages,6 tables, 4 figures |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.24478 [cs.CL] |
| (or arXiv:2610.24478v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.24478 arXiv-issued DOI via DataCite |
Submission history
From: Yunhan Li [view email]
[v1]
Sat, 17 Oct 2026 05:49:46 UTC (2,390 KB)
[v2]
Thu, 22 Oct 2026 04:11:50 UTC (2,390 KB)
来源：arXiv cs.CL · arxiv.org