Abstract

Organizations globally recognize the significance of sustainability disclosures as a critical business metric. These disclosures are made public through various channels, with many organizations publishing dedicated sustainability reports to showcase achievements across the Environment, Social, and Governance (ESG) pillars. The lack of a standardized format has made the extraction of sustainability parameters from these reports a prominent research area in Natural Language Processing (NLP). With advancements in language modelling and the availability of open-source Large Language Models (LLMs), the field of Information Systems is increasingly focusing on Information Extraction and Retrieval using these technologies. This work introduces a comprehensive system that leverages LLMs to extract and verify sustainability disclosures from organizational reports. The methodology systematically locates ESG disclosures and assesses their compliance with established reporting frameworks using LLM-based modules. Our experiments reveal variations in the quality of disclosures across sectors, with some industries closely adhering to established frameworks while others demonstrate gaps in fulfilling key requirements. Additionally, many organizations report non-substantive sustainability parameters that lack meaningful insights. These findings offer valuable input for downstream decision-making related to sustainability analysis.

Download

Share

COinS