← 論文一覧に戻る

人工知能とESG開示:象徴的・実質的サステナビリティ・コミュニケーション識別における大規模言語モデルの評価

Artificial Intelligence and ESG Disclosure: Evaluating Large Language Models for Identifying Symbolic and Substantive Sustainability Communication (原題)

Anuj Pal

Crossrefプレプリント2026-01-01#AI×ESG経営インパクト: 調達リスク対象セクター: cross_sector
DOI: 10.2139/ssrn.7160759
原典: https://doi.org/10.2139/ssrn.7160759

🤖 gxceed AI 要約

日本語

本研究は、ESG開示における「象徴的(symbolic)」と「実質的(substantive)」なサステナビリティ・コミュニケーションの区別を、大規模言語モデル(GPT-5.5)が支援できるかを検証した。5社の多国籍企業による35件のESG開示を対象に、研究者参加型の質的内容分析を実施。GPT-5.5は明確な象徴的・実質的開示を誤分類なく識別し、不一致は主に両特性を併せ持つ曖昧な開示に集中した。これを踏まえ、Symbolic・Mixed・Substantiveの3分類からなるOEDCF(Operational ESG Disclosure Coding Framework)を提案し、AIが専門家判断を代替ではなく補完することを示した。

English

This study tests whether GPT-5.5 can help researchers distinguish symbolic from substantive sustainability communication in ESG disclosures. Using a researcher-in-the-loop qualitative content analysis of 35 disclosures from five multinational firms, the model cleanly separated clearly symbolic and substantive cases with no direct misclassification; disagreements clustered on mixed disclosures. The authors introduce the Operational ESG Disclosure Coding Framework (OEDCF) with Symbolic, Mixed, and Substantive codes, showing LLMs can augment rather than replace expert judgement and offering a transparent, replicable coding protocol.

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

SSBJ基準や有価証券報告書でのサステナビリティ開示が拡大する日本では、開示の「形式化(象徴的)」対「実質性」を見極める評価手法への需要が高い。OEDCFのような再現可能なコーディング枠組みは、国内企業の統合報告書や投資家向けESG評価の質的検証に応用できる。

In the global GX context

As ISSB/CSRD and SEC climate rules raise disclosure volume, regulators and investors increasingly worry about symbolic versus substantive reporting and greenwashing. This paper contributes a replicable, AI-assisted coding framework that can strengthen disclosure-quality assessment and greenwashing detection in global disclosure scholarship.

👥 読者別の含意

🔬研究者:LLMを質的ESG開示分析に組み込む際の検証設計とMixed分類の必要性を示す実証的知見を提供する。

🏢実務担当者:自社開示が象徴的と見なされないか、実質的エビデンスを伴う記述になっているかを点検する枠組みとして活用できる。

🏛政策担当者:開示品質・グリーンウォッシュ監視においてAI支援の質的評価手法を規制・審査に取り込む際の参考になる。

📄 Abstract(原文)

<div> Environmental, Social, and Governance (ESG) disclosure has become an in- </div> <div> creasingly important component of corporate reporting, creating a growing need </div> <div> for rigorous evaluation of disclosure quality. Although the distinction between </div> <div> symbolic and substantive sustainability communication provides an established an- </div> <div> alytical lens for assessing ESG disclosures, consistently applying this distinction </div> <div> during qualitative content analysis remains challenging, particularly when disclo- </div> <div> sures combine aspirational commitments with evidence of implemented sustainabil- </div> <div> ity practices. Recent advances in artificial intelligence (AI), particularly large lan- </div> <div> guage models (LLMs), offer opportunities to support qualitative research; however, </div> <div> their application to qualitative ESG disclosure analysis remains underexplored. </div> <div> This study investigates whether AI can assist researchers in distinguishing sym- </div> <div> bolic and substantive sustainability communication during qualitative ESG disclo- </div> <div> sure analysis. Using a researcher-in-the-loop qualitative content analysis approach, </div> <div> GPT-5.5 was evaluated on 35 ESG disclosures from five multinational companies. </div> <div> Based on this evaluation, the study introduces the Operational ESG Disclosure </div> <div> Coding Framework (OEDCF), which operationalises qualitative ESG disclosure </div> <div> analysis through three operational coding classifications: Symbolic, Mixed, and </div> <div> Substantive. </div> <div> The findings show that GPT-5.5 successfully distinguished clearly symbolic and </div> <div> substantive disclosures, with no direct misclassification between these two classifica- </div> <div> tions. Most coding disagreements occurred for disclosures exhibiting characteristics </div> <div> associated with both communication styles, supporting the inclusion of the Mixed </div> <div> operational coding classification and reinforcing the importance of researcher val- </div> <div> idation for ambiguous cases. The OEDCF provides a transparent and replicable </div> <div> coding framework that improves the consistency and systematic evaluation of qual- </div> <div> itative ESG disclosure analysis. </div> <div> This study contributes to the emerging literature on AI-assisted qualitative </div> <div> ESG disclosure analysis by demonstrating how large language models can augment, </div> <div> rather than replace, expert qualitative judgement while providing a structured op- </div> <div> erational framework for ESG disclosure coding. </div>

🔗 Provenance — このレコードを発見したソース

🔔 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。