KG-ESG: Explainable Real-Time ESG News Intelligence with Knowledge Graphs and GraphRAG
KG-ESG: 知識グラフとGraphRAGによる説明可能なリアルタイムESGニュースインテリジェンス (AI 翻訳)
Omar Nassar, Fahimeh Jafari, David King
🤖 gxceed AI 要約
日本語
本論文は、KG-ESGというAI駆動の概念実証を提案し、公開ESGニュースを構造化された知識グラフに変換する。FTSE100/250およびASX200企業を対象に、NER、感情分析、ESG分類を適用し、1,098件のニュース記事を処理。DistilBERTの感情精度66%、ゼロショット分類のESG精度60%とプロトタイプ段階だが、GraphRAGによる説明可能性と透明度の高い意思決定支援を実現する。
English
This paper presents KG-ESG, an AI-driven proof-of-concept that converts public ESG news into structured, queryable knowledge graphs. It processes 1,098 news articles for 450 FTSE 100/250 and ASX 200 companies using NER, sentiment analysis, and ESG classification. DistilBERT achieves 66% sentiment accuracy and zero-shot ESG classification 60%, indicating prototype-stage performance. The system includes a GraphRAG layer for transparent, traceable evidence path queries.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本企業への直接適用はないが、KGとGraphRAGを組み合わせたESGニュースインテリジェンス手法は、日本の有報や統合報告書のテキスト分析にも応用可能。SSBJ開示の裏付け情報としての活用が期待される。
In the global GX context
This work responds to the global demand for real-time, evidence-linked ESG data beyond annual reports. Its GraphRAG approach offers a transparent alternative to opaque ESG ratings, potentially supporting ISSB and CSRD-aligned narrative reporting.
👥 読者別の含意
🔬研究者:The paper demonstrates an applied architecture combining knowledge graphs, NLP, and GraphRAG for ESG intelligence, offering a baseline for further research in explainable AI-driven ESG analysis.
🏢実務担当者:The system provides a prototype for corporate sustainability teams to monitor real-time ESG news and gather traceable evidence for stakeholder reporting.
🏛政策担当者:The transparent, evidence-based approach could inform regulators seeking to improve ESG data quality and combat greenwashing.
📄 Abstract（原文）
Environmental, Social and Governance (ESG) intelligence is increasingly used by investors, regulators, auditors and corporations, yet many ESG assessments remain dependent on static disclosures, delayed ratings and opaque scoring methodologies. This paper presents KG-ESG, an implemented AI-driven proof-of-concept that converts public ESG news into structured, queryable and evidence-linked knowledge graph intelligence. The system constructs a company registry for FTSE 100, FTSE 250 and ASX 200 firms, retrieves ESG-relevant news from public sources, applies named entity recognition, sentiment analysis and ESG-domain classification, and stores the enriched evidence in a Neo4j property graph. The implementation processed 550 registered companies and produced a corpus of 1,098 news articles for 450 companies, covering 93 unique sectors grouped into 29 broader categories. Preliminary validation results indicate 66.00% sentiment accuracy for DistilBERT and 60.00% ESG-domain accuracy for zero-shot classification; these results are interpreted as prototype-level evidence rather than final benchmark performance. The system further includes a schema-aware GraphRAG layer that translates natural-language questions into graph queries and returns evidence paths rather than unsupported summaries. The contribution is therefore not a verified ESG-rating mechanism, but a transparent semantic infrastructure for media-derived ESG evidence collection, comparative analysis and traceable decision support.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.1109/isades69945.2026.11608094first seen 2026-07-25 05:56:30
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。