cs.DB提交于 2026-10-19已译

基于迭代自优化的 NL2SQL

NL2SQL with Iterative Self-Refinement

Anjali Vemula · Akshat Dasula

中文摘要

弥合非技术用户意图与机器逻辑间的差距是数据访问的关键挑战，但大语言模型（LLM）常出现模式幻觉、语法错误或意图偏差，这在保险等高精度领域尤为致命。该研究提出一种多智能体、角色驱动的 NL2SQL 框架，整合领域相关性检测、基于模式的知识构建、KPI 感知的 SQL 合成、利用 sqlite 执行反馈的语法自修复，以及通过 SQL 到自然语言逆向工程进行的语义一致性检查。在包含 50 条 HuggingFace 数据集查询和 10 条自定义领域查询的 60 条测试集上，系统使用 Mistral 模型，通过七次协调的 LLM 调用实现了 90% 的正确率。相比直接提示 LLM，该方法显著降低了查询幻觉并提升了语义保真度，证明了其在高风险环境中提升 NL2SQL 管道可靠性的潜力。

关键要点

01针对 LLM 在保险等高要求领域生成 SQL 时存在的幻觉、语法错误及意图偏差问题。
02构建多智能体框架，结合领域检测、模式知识构建、KPI 感知合成及基于执行反馈的自修复。
03在 60 条查询测试中，使用 Mistral 模型经七次协调调用，实现了 90% 的查询正确率。
04相比直接提示 LLM，该方法显著降低了查询幻觉并提升了语义保真度。

解读

尚无解读。

原始英文摘要

arXiv:2610.22968v1 Announce Type: cross Abstract: Bridging the gap between human intent and machine logic remains a critical challenge in data access for non-technical users, yet Large Language Models (LLMs) remain unreliable: they hallucinate schema elements, generate syntactically invalid queries, or fail to capture user intent. These limitations are especially critical in domains such as insurance, where KPIs, regulatory constraints, and legacy relational systems demand high precision. We introduce a multi-agent, persona-driven NL2SQL framework that combines (1) domain relevance detection, (2) schema-grounded knowledge construction, (3) KPI-aware SQL synthesis, (4) syntax-level self-repair using sqlite execution feedback, and (5) semantic consistency checking via SQL-to-NL reverse engineering. Our system significantly reduces query hallucination and improves semantic fidelity compared to direct LLM prompting. Evaluated on 60 queries (50 from a HuggingFace text-to-SQL dataset and 10 custom domain queries), the system achieves 90% correctness using Mistral across seven coordinated LLM calls showing that this approach can substantially increase the reliability of NL2SQL pipelines in high-stakes environments.