KGMistral Approach for Enhancing QA with Knowledge Graphs

This paper presents a technical report on KGMistral, a novel question-answering (QA) approach that integrates domain-specific Knowledge Graphs (KGs) to enhance the performance of Large Language Models (LLMs), specifically the Mistral model. The study addresses the limitations of LLMs in generating accurate answers, particularly in specialized domains like materials science and engineering, by leveraging the Retrieval Augmented Generation (RAG) framework. The report outlines the architecture of KGMistral, detailing the processes of entity and relation extraction, similarity matching, and the use of SPARQL queries to retrieve relevant triples from KGs. Experimental results demonstrate that KGMistral significantly improves QA performance by utilizing external knowledge sources. The findings contribute to the understanding of how integrating KGs can mitigate issues such as hallucinations in LLMs, thereby providing a more reliable framework for question answering in specialized fields.