ESG 메트릭 지식 그래프 자동 구축을 위한 온톨로지 기반 프레임워크

OntoMetric: An Ontology-Driven LLM-Assisted Framework for Automated ESG Metric Knowledge Graph Generation Mingqin Yu∗ mingqin.yu@unsw.edu.au School of Computer Science and Engineering, University of New South Wales Sydney, Australia Fethi Rabhi f.rabhi@unsw.edu.au School of Computer Science and Engineering, University of New South Wales Sydney, Australia Boming Xia boming.xia@adelaide.edu.au Faculty of Sciences, Engineering and Technology, The University of Adelaide Adelaide, Australia Zhengyi Yang zhengyi.yang@unsw.edu.au School of Computer Science and Engineering, University of New South Wales Sydney, Australia Felix Tan f.tan@unsw.edu.au School of Information Systems and Technology Management, University of New South Wales Sydney, Australia Qinghua Lu qinghua.lu@data61.csiro.au CSIRO’s Data61 Sydney, Australia Abstract Environmental, Social, and Governance (ESG) metric knowledge is inherently structured, connecting industries, reporting frame- works, metric categories, metrics, and calculation models through compositional dependencies, yet in practice this structure remains embedded implicitly in regulatory documents such as SASB, TCFD, and IFRS S2 and rarely exists as an explicit, governed, or machine- actionable artefact. Existing ESG ontologies define formal schemas but do not address scalable population and governance from au- thoritative regulatory sources, while unconstrained large language model (LLM) extraction frequently produces semantically incorrect entities, hallucinated relationships, and structurally invalid graphs. OntoMetric is an ontology-guided framework for the automated construction and governance of ESG metric knowledge graphs from regulatory documents that operationalises the ESG Metric Knowl- edge Graph (ESGMKG) ontology as a first-class constraint embed- ded directly into the extraction and population process. The frame- work integrates structure-aware segmentation, ontology-constrained LLM extraction enriched with semantic fields and deterministic identifiers, and two-phase validation combining semantic type verification with rule-based schema checking, while preserving segment-level and page-level provenance to ensure traceability to regulatory source text. Evaluation on five ESG regulatory standards shows that ontology-guided extraction achieves 65–90% semantic accuracy and over 80% schema compliance, compared with 3–10% for unconstrained baseline extraction, and yields stable cost effi- ciency with a cost per validated entity of $0.01–$0.02 and a 48× efficiency improvement over baseline. Keywords Ontology-Guided LLM Extraction, ESG Knowledge Graphs, Two- Phase Validation, Provenance Preservation, Regulatory Knowledge Engineering, AI-Ready Knowledge Representation ∗Corresponding author. 1 Introduction Environmental, Social, and Governance (ESG) metrics constitute a structured body of domain knowledge that specifies what must be measured, how values are computed, which units apply, and how individual indicators depend on one another. Beyond simple nu- merical values, ESG metric knowledge includes formal definitions, scope and boundary conditions, disaggregation rules, calculation models, and compositional dependencies between metrics and their input variables. Collectively, these elements form an implicit ESG metric knowledge graph that connects industries, reporting con- texts, metric categories, metrics, and computational models into a coherent semantic structure. In practice, however, this ESG metric knowledge graph does not exist as an explicit, governed, or machine-actionable artefact. Instead, metric definitions, calculation logic, and inter-metric de- pendencies are embedded implicitly in regulatory documents and reporting artefacts. Where ESG ontologies have been proposed, they typically provide only formal schemas or high-level concept hierarchies, without populated instances or traceable links to au- thoritative metric definitions. As a result, the construction and maintenance of ESG metric knowledge graphs remains a largely manual, expert-driven process. Several efforts have sought to formalise ESG knowledge using ontologies and semantic models. Notably, we previously proposed the ESG Metric Knowledge Graph (ESGMKG), an ontology-driven architecture that defines the core entity types, relationships, and compositional structures required to represent ESG metric knowl- edge, including industries, reporting frameworks, metric categories, metrics, and calculation models [16]. However, ESGMKG and re- lated ontologies focus primarily on schema definition rather than scalable population and governance. They do not address how ESG metric knowledge can be constructed automatically from regulatory sources or how provenance can be preserved at scale. In real-world settings, ESG metric knowledge is derived from multiple sources, among which regulatory and quasi-regulatory arXiv:2512.01289v2 [cs.AI] 26 Jan 2026 Mingqin Yu, Fethi Rabhi, Boming Xia, Zhengyi Yang, Felix Tan, and Qi

…(Full text truncated)…

This content is AI-processed based on ArXiv data.