ESG管理・報告の自動化に向けたAI活用オントロジー駆動型指標システム

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:ESG指標の意味論的モデリングとLLM抽出を統合する設計科学の枠組みとして、開示インフラ研究の基盤になりうる。

🏢実務担当者:指標定義・算定ロジック・出典を構造化し、規制改定に追随できる報告システム設計の指針として活用できる。

🏛政策担当者:規制文書の機械可読化と指標の追跡可能性確保が、開示制度の相互運用性と保証負担軽減に資する点を示唆する。

Environmental, Social, and Governance (ESG) considerations have become central to contemporary regulation, finance, and corporate decision-making. As ESG requirements have shifted from voluntary disclosure to formal regulatory obligations, ESG metrics have emerged as the primary mechanism through which sustainability concepts are operationalised within software systems. In practice, however, ESG metrics are typically implemented as flat data variables embedded in spreadsheets, reporting tools, and procedural code. Meanwhile, their definitions, calculation logic, derivation dependencies, and provenance remain implicit in regulatory documents. This fragmentation limits transparency, traceability, governance, and adaptability as reporting standards evolve. This thesis addresses this challenge by reframing ESG metrics as a problem of domain knowledge representation and system design, rather than solely data management or regulatory compliance. Adopting a Design Science Research methodology, the thesis develops three complementary artefacts that collectively address key stages of the ESG metric system lifecycle: • ESG Metric Knowledge Graph (ESGMKG): a formal, metric-centric semantic representation that models ESG metrics, calculation models, derivation dependencies, reporting frameworks, data inputs, and provenance relationships as first-class conceptual entities, independent of specific datasets or ESG data providers, enabling explicit governance and traceability. • Ontology-driven ESG Metric System for Reporting: a modular software system that operationalises ESGMKG for executable ESG reporting, demonstrating how ontology-defined metric knowledge supports metric discovery, dependency navigation, and computation, and enables the assembly of computed metrics into structured, provenance-aware ESG reports, while preserving links to metric definitions, calculation logic, and underlying data sources. • An Ontology-Guided Framework for Automated ESG Metric Knowledge Construction: an ontology-guided approach for scalable construction and maintenance of ESG metric knowledge from unstructured regulatory documents, integrating structure-aware document segmentation, ontology-guided large language model extraction, and a two-phase validation process to preserve semantic fidelity and provenance. By separating semantic specification, system operationalisation, and automated knowledge construction into clearly defined artefacts, this thesis demonstrates how ontology-driven representation and controlled AI-assisted automation can support transparent, governable, and adaptable ESG metric systems. The work provides a foundational contribution to the design of software systems for ESG reporting and sustainability analytics in evolving regulatory environments.

• openaire https://doi.org/10.26190/unsworks/32592first seen 2026-11-05 04:43:38