Detecting Issues Related to Environmental, Social, and Corporate Governance Using SEC-BERT

Responsible investors seek information related to Environmental, Social, and Corporate Governance (ESG) while making investment decisions. However, reading through news articles for figuring out ESG-related issues is a tedious task. In this paper, we show how Natural language Processing (NLP) can be used to detect ESG-related issues from financial news. Experiments with different variants of Bidirectional Encoder Representations from Transformers (BERT) models reveal that finance domain-specific BERT (i.e., SEC-BERT), pre-fine-tuned on news articles related to ESG and subsequently fine-tuned for classification, performs the best in detecting ESG-related issues. Moreover, we developed a user-friendly tool—ESG Issue Detector (EID) to empower investors. We shall open-source the final model and the EID tool after the acceptance of this manuscript.