Large language model-based knowledge graph construction of unstructured accident reports to improve construction safety

Tools

Liu, Qiong and Luo, Xiaowei; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Large language model-based knowledge graph construction of unstructured accident reports to improve construction safety. In: EG-ICE 2025. University of Strathclyde Publishing, GBR, pp. 536-546. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093269)

[thumbnail of Liu-Luo-EG-ICE-2025-Large-language-model-based-knowledge-graph-construction]
Preview
 Text. Filename: Liu-Luo-EG-ICE-2025-Large-language-model-based-knowledge-graph-construction.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (4MB)| Preview

Abstract

To effectively capture and represent safety knowledge, this study proposes a novel framework for automated construction accident knowledge graph generation using large language models (LLMs). First, an improved Accimap model is used to determine the ontology framework for the knowledge graph. Then, by leveraging iterative chaining prompts based on Chain-of-Thought and Self-consistency principles, the method achieves zero-label extraction of named entities recognition (NER) and relation extraction (RE) from unstructured text. Finally, the extracted triples are stored to provide visual knowledge query services. Experimental results show that the proposed method achieves an accuracy of 0.97 in NER task and 0.90 in RE task under zero-shot conditions. The proposed framework eliminates the need for external resources or human expertise and offers a novel knowledge management approach for construction site safety management.

  • Item type:Book Section
    ID code:93269
    Dates:
    Date
    Event
    1 July 2025
    Published
    9 June 2025
    Accepted
    Subjects:Fine Arts > Architecture
    Department:Faculty of Engineering > Architecture
    Depositing user: Pure Administrator
    Date deposited:27 Jun 2025 10:53
    Last modified:20 May 2026 00:37
    URI:https://strathprints.strath.ac.uk/id/eprint/93269
CORE (COnnecting REpositories)