Enhancing Retrieval for ESGLLM via ESG-CID -- A Disclosure Content Index Finetuning Dataset for Mapping GRI and ESRS

2025-03-10

AI Summary

Retrieval-augmented generation (RAG) systems for automated ESG reporting face a critical bottleneck: the scarcity of labeled training data for aligning disclosures with sustainability standards. Method: This paper introduces ESG-CID—the first benchmark dataset for GRI–ESRS alignment—leveraging intrinsic disclosure indices within ESG reports as weak supervision signals. We systematically mine and structure these indices to automatically construct mappings between standard clauses and report passages. Our retrieval framework integrates fine-tuned BERT embeddings, LLM-as-a-judge quality filtering, and temporal segmentation–based evaluation to ensure robust generalization across frameworks (GRI→ESRS) and heterogeneous reporting styles. Contribution/Results: Experiments demonstrate that our method significantly outperforms commercial and state-of-the-art open-source embedding models on ESG-CID, achieving an 18.7% absolute gain in cross-style retrieval accuracy. This validates the effectiveness and scalability of weakly supervised construction for high-quality, standards-aligned ESG retrieval data.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: Information Extraction
• Humans and AI: Crowd Sourcing and Human Computation

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Web data integration and cleaning

Abstract

Climate change has intensified the need for transparency and accountability in organizational practices, making Environmental, Social, and Governance (ESG) reporting increasingly crucial. Frameworks like the Global Reporting Initiative (GRI) and the new European Sustainability Reporting Standards (ESRS) aim to standardize ESG reporting, yet generating comprehensive reports remains challenging due to the considerable length of ESG documents and variability in company reporting styles. To facilitate ESG report automation, Retrieval-Augmented Generation (RAG) systems can be employed, but their development is hindered by a lack of labeled data suitable for training retrieval models. In this paper, we leverage an underutilized source of weak supervision -- the disclosure content index found in past ESG reports -- to create a comprehensive dataset, ESG-CID, for both GRI and ESRS standards. By extracting mappings between specific disclosure requirements and corresponding report sections, and refining them using a Large Language Model as a judge, we generate a robust training and evaluation set. We benchmark popular embedding models on this dataset and show that fine-tuning BERT-based models can outperform commercial embeddings and leading public models, even under temporal data splits for cross-report style transfer from GRI to ESRS

Problem

• Addresses lack of labeled data for ESG report automation.
• Creates ESG-CID dataset for GRI and ESRS standards.
• Improves retrieval models for ESG report generation.

Innovation

• Utilizes ESG-CID dataset for GRI and ESRS standards
• Employs Large Language Model for data refinement
• Fine-tunes BERT-based models for superior performance