arXiv cs.AI· Padmaja Jonnalagedda, Yuguang Yao, Xiang Gao, Hilaf Hasson, Kamalika Das·· 14 小时前AI 评分38
可执行 Schema 契约：从自动摄取到多源检索
Executable Schema Contracts: From Automatic Ingestion to Multi-Source Retrieval
AI 导读
一个系统可从原始多源数据中自动发现可执行 schema，并将其作为知识图谱构建与查询检索的共享契约。该系统用封闭字段目录约束 LLM 的 schema 发现、以确定性结构分析推断主键外键与来源层级，查询时由 schema 条件化多工具智能体在结构化查询、图遍历和向量检索间路由。在相同 LLM、数据和评测框架的零样本对比中，该系统在四个 QA benchmark 上优于仅检索和基于分解的基线。
正文 · 原文
Abstract:Real-world data spans tables, documents, and semi-structured files with implicit semantics. Querying this data requires integrating evidence across inconsistent schemas and formats, yet existing approaches either demand costly manual engineering or bypass structure entirely. We present a system that automatically discovers an executable schema from raw multi-source data and uses it as a shared contract for knowledge graph construction and query-time retrieval. A closed-world field catalog constrains LLM-based schema discovery to attested fields; deterministic structural analysis infers identity keys, foreign keys, and source hierarchy; and the resulting schema drives extraction, deduplication, and cross-source linking into a provenance-aware knowledge graph. At query time the schema -- optionally extended via a monotonic protocol -- conditions a multi-tool agent routing retrieval across structured lookup, graph traversal, and vector search, returning grounded answers with traceable citations. In controlled zero-shot comparisons using the same LLM, data, and evaluation harness, the system improves over retrieval-only and decomposition-based baselines across four QA benchmarks, with ablations showing that schema-conditioned routing, structural intelligence, and schema-guided construction each contribute to the gains.
| Comments: | 9 pages, 4 figures, plus supplementary appendix |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.05415 [cs.CL] |
| (or arXiv:2606.05415v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.05415 arXiv-issued DOI via DataCite |
Submission history
From: Saisri Padmaja Jonnalagedda [view email]
[v1]
Wed, 3 Jun 2026 20:28:36 UTC (821 KB)
[v2]
Wed, 28 Oct 2026 22:31:23 UTC (823 KB)
来源：arXiv cs.AI · arxiv.org