SYNTRA：自改进数据合成可跨领域规模化适配嵌入模型
Self-improving data synthesis can scale embedding-model adaptation across domains and tasks
AI 导读
研究者提出 SYNTRA，一种自改进的智能体数据合成框架，可将少量检索示例转化为可复用的合成指令，仅需无标注语料和少量查询-文档对即可推断信息需求与相关性标准。在覆盖 15 个领域的 24 个数据集上，同一合成流程生成的训练数据可提升强嵌入模型，在六个任务上均强于基线，其中四个任务检索效果超过人工校准。在 MS MARCO 上，检索增益在 32 倍的合成数据规模范围内持续增长。
正文 · 原文
Abstract:AI agents require specialised retrieval across domains and tasks, but adapting embedding models is constrained by costly manual training-data design. Here we introduce SYNTRA, a self-improving agentic data synthesis framework that turns a few retrieval examples into reusable synthesis instructions. Given an unlabelled corpus and a few relevant query-document pairs, SYNTRA infers implicit information needs and relevance criteria, then uses feedback on trial outputs to refine instructions for query generation and relevance assessment. Experiments across 24 datasets spanning 15 domains demonstrate that the same synthesis workflow produces effective training data, benefiting even strong embedding models with and without existing labelled training sets. Self-improvement strengthens this supervision on all six evaluated tasks, yielding better retrieval than human calibration on four. The training value persists at larger scales: on MS MARCO, retrieval gains continue across a 32-fold range of synthetic training-data sizes. SYNTRA thus enables scalable embedding-model adaptation with minimal human intervention.
| Comments: | 32 pages, 6 figures; supplementary information included (27 pages). Code: this https URL . Data: this https URL |
| Subjects: | Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.14730 [cs.IR] |
| (or arXiv:2610.14730v3 [cs.IR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.14730 arXiv-issued DOI via DataCite |
Submission history
From: Zucan Lyu [view email]
[v1]
Sat, 10 Oct 2026 08:50:48 UTC (31,557 KB)
[v2]
Thu, 15 Oct 2026 03:14:24 UTC (31,557 KB)
[v3]
Fri, 30 Oct 2026 16:26:03 UTC (31,564 KB)
来源：arXiv cs.AI · arxiv.org