BASE-ESG: ESG Knowledge Bases for Responsible Generative AI

Construction of ESG Knowledge Bases for the Responsible Training of Generative AI Systems from Public Disclosures (BASE-ESG) is a research project dedicated to the construction of structured ESG (Environmental, Social, and Governance) knowledge bases derived from public corporate disclosures, with the goal of enabling the responsible training and evaluation of generative AI systems in the sustainability domain.

As ESG reporting becomes increasingly central to corporate accountability and investment decisions, the volume and complexity of relevant textual disclosures — including sustainability reports, regulatory filings, and stakeholder communications — has grown substantially. This project addresses the need for high-quality, structured knowledge resources that can ground generative AI models in accurate, verifiable ESG information, reducing hallucination and improving reliability in this high-stakes domain.

Objectives

• Extract and structure ESG-relevant information from public corporate disclosures using NLP techniques
• Build annotated datasets and knowledge bases covering environmental, social, and governance dimensions
• Develop and evaluate retrieval-augmented and knowledge-grounded generative AI systems for ESG question answering and analysis
• Promote responsible AI practices in the financial and sustainability sectors

Funding & Support

This project is financed by ICT Itaú, covering the period from January 2026 to December 2027.