Executable Schema Contracts: From Automatic Ingestion to Multi-Source Retrieval
Researchers have developed a system that automatically generates an executable schema from diverse data sources, including tables, documents, and semi-structured files. This schema acts as a shared contract to construct knowledge graphs and facilitate retrieval across these varied formats. The system leverages LLMs for schema discovery, structural analysis for key inference, and a provenance-aware knowledge graph for extraction and linking, ultimately improving question-answering benchmarks. AI
IMPACT Enables more robust and automated knowledge graph construction and question-answering across disparate data sources.