Knowledge engineering with LLMs for asset information management in the built environment

The integration of advanced technologies into asset information management within the built environment addresses challenges related to knowledge management and unstructured data. Recent research emphasises the development of ontologies and knowledge graphs (KGs) in the construction sector, as the synergic integration of KGs and large language models (LLMs). This study proposes a methodology for automating KG generation from domain document corpora leveraging LLMs inference skills, generating Terse RDF Triple Language serialised triples and aligning extracted entities and relationships with domain ontologies. The model is asked to perform knowledge engineering tasks through a guided prompt, based on a few-shot prompting strategy including use-case relevant exemplars. Alignment with ontological semantics, as instructed through prompting, is evaluated, as well as syntactic consistency of extracted entities and relationships. The study compares the performance of three inferencing scenarios (providing input as full-text, paragraphs, or individual sentences), highlighting limitations and possible future improvements of the system. Initial applications demonstrate the pipeline’s effectiveness in creating KGs that represent asset management needs, highlighting the potential of LLMs to facilitate knowledge engineering tasks and paving the way for improved data structuring and accessibility in the built environment.