KG-RAG4SM retrieves knowledge-graph subgraphs for schema matching
Original title: Knowledge Graph-based Retrieval-Augmented Generation for Schema Matching
Authors report precision gains of 39.79% on MIMIC and 69.20% on Synthea versus named prior methods, without retraining.
KG-RAG4SM uses LLM-based, vector-based, graph-traversal, and hybrid retrieval with reranking to extract relevant subgraphs from external knowledge graphs for schema matching. The authors report 39.79% higher precision and 30.56% higher F1 than LLM-based methods such as Jellyfish-8B on MIMIC, and 69.20% higher precision and 21.99% higher F1 than PLM-based methods such as SMAT on Synthea. An instruction-tuning knowledge exclusion component is intended to keep end-to-end matching truthful when KG context is weakly relevant.
Unconfirmed: Precision and F1 improvements are the authors' reported figures, not independently confirmed here.
- The method does not require retraining, according to the paper.
- It is tested from commonsense knowledge graphs to domain-specific graphs.
- Authors say case studies show reduced LLM hallucination in schema matching.
- Comparisons name Jellyfish-8B on MIMIC and SMAT on Synthea.
- Products
- KG-RAG4SM
- AI models
- Jellyfish-8B
From the original article
1 min read at huggingface.co
Traditional similarity-based schema matching methods fail to resolve semantic ambiguities and conflicts in complex, domain-specific mappings due to a lack of commonsense and domain-specific knowledge. In parallel, large language models (LLMs) without sufficient background knowledge tend to hallucinate, further complicating schema matching.
By Chuangtao Ma, Sriom Chakrabarti, Arijit Khan · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.