Abstract
This paper introduces SemGraphRAG, a hybrid retrieval-augmented generation (RAG) solution that integrates Semantic Knowledge Graphs (KGs) with large language models (LLMs). SemGraphRAG offers a generic and extensible mechanism for incorporating structured semantic relationships into RAG pipelines, enabling interoperability with diverse knowledge bases and ontologies. We demonstrate the approach in a scenario focused on expert competence retrieval, comparing traditional text-chunking methods (Naive RAG) with our semantic graphbased strategy. An evaluation on 40 expert queries shows that semantic graph integration significantly enhances retrieval and answer quality, with SemGraphRAG achieving higher precision (0.75 vs. 0.59), recall (0.67 vs. 0.60), and accuracy (0.95 vs. 0.90) compared to Naive RAG. These findings underscore the benefits of semantic enrichment for LLM-based information systems and highlight SemGraphRAG's potential to advance knowledgedriven applications in the Semantic Web.