A Multi-Agent System for Semantic Mapping of Relational Data to Knowledge Graphs

TL;DR

The paper addresses interoperability challenges from siloed relational databases by proposing a semantic layer that maps tables and columns to Schema.org concepts using a multi-agent LLM system to construct a unified knowledge graph. It introduces a graph-vector store to support retrieval-augmented semantic mapping and assigns dedicated agents for mapping, relation discovery, and validation, operating in a table-wise pipeline. The approach achieves high mapping accuracy across domains in the Spider benchmark (up to $93.54\%$ in Apartments and $78.72\%$ in Retail) and demonstrates promising cross-domain applicability for enterprise data integration. This work reduces manual data wrangling and enhances interoperability by leveraging existing vocabularies and a scalable, automated reasoning workflow.

Abstract

Enterprises often maintain multiple databases for storing critical business data in siloed systems, resulting in inefficiencies and challenges with data interoperability. A key to overcoming these challenges lies in integrating disparate data sources, enabling businesses to unlock the full potential of their data. Our work presents a novel approach for integrating multiple databases using knowledge graphs, focusing on the application of large language models as semantic agents for mapping and connecting structured data across systems by leveraging existing vocabularies. The proposed methodology introduces a semantic layer above tables in relational databases, utilizing a system comprising multiple LLM agents that map tables and columns to Schema.org terms. Our approach achieves a mapping accuracy of over 90% in multiple domains.

Figures (2)

• Figure 1: Graph - Vector store construction with ontology sub-graphs
• Figure 2: Data flow of the semantic multi-agent system