Knowledge Graph-Enhanced LLMs for Unified Climate-Aware Risk Management in Banking: The Cognitive Bank Architecture
銀行における統一的な気候認識リスク管理のための知識グラフ強化LLM：コグニティブバンクアーキテクチャ (AI 翻訳)
Rohit Nimmala, Gajendra Babu Thokala, Jagrut Nimmala, Milan Parikh
🤖 gxceed AI 要約
日本語
本論文は、気候リスクが信用・市場・オペレーショナルリスクに伝播する問題に対し、知識グラフとLLMを統合した「Cognitive Bank」アーキテクチャを提案。FIBOベースの気候金融KG、グラフ制約推論によるLLM強化、連合学習と差分プライバシーを組み合わせ、TCFD準拠の報告を目指す。設計目標として信用リスクで92%以上の均衡精度、AML検出で20%以上の改善を掲げる。
English
This paper proposes the Cognitive Bank architecture that integrates knowledge graphs, LLMs, and federated learning to unify climate risk assessment across credit, market, and operational risks. It uses a FIBO-based climate-financial KG, Graph-Constrained Reasoning to reduce hallucinations, and differential privacy. Design targets include 92%+ balanced accuracy on climate-adjusted credit risk and 20%+ improvement on cross-institutional AML detection, with full TCFD-aligned reporting.
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
日本ではSSBJ基準や有報での気候関連開示が進む中、銀行のリスク管理統合は実務課題。本アーキテクチャは知識グラフとLLMを活用し、規制サイロを解消する点で、邦銀のTCFD対応や気候リスク管理高度化に示唆を与える。
In the global GX context
This paper addresses the BCBS-identified silo problem in climate risk management by proposing a unified architecture. It is highly relevant to global TCFD/ISSB disclosure trends and the growing adoption of AI in banking risk systems, offering a blueprint for integrated climate-financial risk analytics.
👥 読者別の含意
🔬研究者:Provides a novel integration of KGs, LLMs, and federated learning for climate risk, offering a foundation for empirical validation.
🏢実務担当者:Banks can use this architecture to design unified climate risk systems that meet TCFD requirements and improve cross-risk analytics.
🏛政策担当者:Regulators (e.g., BCBS) should note this as a potential solution to the supervisory identified problem of siloed risk management.
📄 Abstract（原文）
Climate risk propagates across credit, market, and operational risk domains through interconnected transmission channels. Existing banking risk systems treat these domains in regulatory silos, as identified by the Basel Committee on Banking Supervision (BCBS) in d517 (2021) and d532 (2022). No existing framework unifies climate risk assessment across all three risk types using knowledge graphs (KGs) and large language models (LLMs). This paper presents the Cognitive Bank, a new architecture with three integrated layers: (1) a climate-financial KG built on the Financial Industry Business Ontology (FIBO) (Bennett, 2013); (2) a knowledge-augmented LLM reasoning engine employing Graph-Constrained Reasoning (GCR) (Luo et al., 2025, ICML) to reduce hallucinations; and (3) federated learning via FedAvg (McMahan et al., 2017) with differential privacy (Abadi et al., 2016). The architecture sets design targets of 92%+ balanced accuracy on climate-adjusted credit risk, informed by Mitra et al. (2024), and 20%+ improvement on cross-institutional anti-money laundering detection, informed by Suzumura et al. (2019), along with full Task Force on Climate-related Financial Disclosures (TCFD)-aligned reporting. To our knowledge, this is the first architecture proposal that brings together knowledge graphs, LLMs, federated learning, climate finance, and banking risk management.
🔗 Provenance — このレコードを発見したソース
- semanticscholar https://doi.org/10.1109/icaiset66439.2026.11541423first seen 2026-06-05 05:22:34
🔔 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。