Extraction SDGs-related sentences from Sustainability Reports using BERT and ChatGPT.

The Sustainable Development Goals (SDGs) call for the voluntary efforts of governments, businesses, and societies worldwide to ensure a sustainable society that leaves no one behind on Earth. In addition, the Principles for Responsible Investment (PRI) have been established for institutional investors, incorporating environmental, social, and governance (ESG) perspectives into investment decisions. Institutional investors have a responsibility to not only promote their own SDGs but also to understand the progress of ESG and Sakaji SDGs within their investment universe. On the other hand, at present, disclosures regarding ESG and SDGs initiatives in Japanese companies primarily exist in the form of PDF-based sustainability reports or similar documents. Consequently, ESG investment managers must visually assess the status of each company’s initiatives, resulting in inefficiencies. In this paper, we propose a methodology for constructing a model that automatically extracts SDG-related sentences from PDF-based information, such as sustainability reports, by fine-tuning the pre-trained BERT model. Furthermore, we used ChatGPT in part for generating training data and evaluated its effectiveness in the process.