HelpsKG: Semi-Automatic Construction of Housing Exploration LPG GDBMS With Knowledge Graph

초록

Knowledge graphs for housing exploration require both high expressivity and query efficiency. This paper presents HelpsKG, a semi-automatic framework that adopts the Labeled Property Graph (LPG) model instead of the Resource Description Framework (RDF), leverages large language models (LLMs), and integrates ontology engineering practices. HelpsKG comprises five phases: (1) Specification, which systematically generates and refines Buyer Profiles and Competency Questions (CQs); (2) Ontology Extraction, which derives core concepts and relations from the CQs; (3) Ontology Completion, which enriches the ontology and maps it to an LPG Validating Schema; (4) LPG Modeling, which implements the ontology in a Neo4j Graph DBMS; and (5) KG Instance Construction, which collects, cleans, and ingests instance data. By decomposing end-to-end KG construction into LLM-assisted, human-in-the-loop tasks, HelpsKG substantially reduces the required effort. We demonstrate effectiveness through comprehensive query-answering evaluations on a housing ontology for Seoul, Korea.

키워드