An Automated LLM-based Pipeline for Asset-Level Database Creation to Assess Deforestation Impact

TL;DR

The paper tackles the challenge of producing asset-level deforestation impact data under the EU Deforestation Regulation by building an end-to-end pipeline that extracts, structures, cleans, and validates asset information from SEC EDGAR filings using LLMs. It introduces Instructional, Role-Based, Zero-Shot Chain-of-Thought (IRZ-CoT) prompting and Retrieval-Augmented Validation (RAV) to enhance extraction accuracy and data reliability, validated against LSEG databases and real-time web sources. Key contributions include a robust data-processing workflow, a ground-truth benchmark, a comparative analysis of LLMs versus traditional NER, a three-step data cleaning process, and a dual-validation framework that significantly improves coverage for asset-level environmental data. The methodology enables scalable regulatory compliance, CSR, and ESG analyses with broad sectoral applicability, while acknowledging limitations and outlining avenues for future enhancements.

Abstract

The European Union Deforestation Regulation (EUDR) requires companies to prove their products do not contribute to deforestation, creating a critical demand for precise, asset-level environmental impact data. Current databases lack the necessary detail, relying heavily on broad financial metrics and manual data collection, which limits regulatory compliance and accurate environmental modeling. This study presents an automated, end-to-end data extraction pipeline that uses LLMs to create, clean, and validate structured databases, specifically targeting sectors with a high risk of deforestation. The pipeline introduces Instructional, Role-Based, Zero-Shot Chain-of-Thought (IRZ-CoT) prompting to enhance data extraction accuracy and a Retrieval-Augmented Validation (RAV) process that integrates real-time web searches for improved data reliability. Applied to SEC EDGAR filings in the Mining, Oil & Gas, and Utilities sectors, the pipeline demonstrates significant improvements over traditional zero-shot prompting approaches, particularly in extraction accuracy and validation coverage. This work advances NLP-driven automation for regulatory compliance, CSR (Corporate Social Responsibility), and ESG, with broad sectoral applicability.

Figures (13)

• Figure 1: System design of end-to-end LLM-based pipeline designed to handle systematic data extraction, structured database creation, cleaning and validation, and the improvement module to increase validation coverage.
• Figure 2: Comparison of different prompt engineering techniques across various evaluation metrics
• Figure 3: Partial match scores from the LSEG database validation. The dotted lines separate the sectors, where the sectors are mining, oil & gas, and utilities, respectively.
• Figure 4: A feedback loop linking physical asset database creation with improved compliance and ESG initiatives, driving continuous refinement.
• Figure 5: Number of chunks generated per document for each company across the three sectors.