Intelligent Knowledge Graph Question Answering Method for Health and Safety Hazard Management Using Large Language Models

Abstract: Construction safety knowledge is often scattered across unstructured and semi-structured sources, complicating retrieval and reasoning. This paper introduces a novel Knowledge Graph Question Answering (KGQA) method leveraging Large Language Models (LLMs) for intelligent QA over safety hazard knowledge. The approach integrates an LLM-based assistant for natural language understanding (NLU), converting natural language queries (NLQs) into structured queries that retrieve information from a domain-specific safety KG. An NLQ dataset is constructed, and QA performance is benchmarked across GPT-4o, DeepSeek-v3, and Claude-3.5 to evaluate our method. Experimental results demonstrate that our method achieves high accuracy and efficiency in safety knowledge retrieval.