[3K4-GS-10] AI application: text analysis

[3K4-GS-10-01] Automatic extraction of ESG-related information from integrated reports

〇Miyu Kodama¹, Hiroyuki Sakai¹, Kengo Enami¹, Kaito Takano², Kei Nakagawa² (1. SEIKEI University, 2. Nomura Asset Management Co.,Ltd.)

Keywords:ESG, integrated report, information extraction, text mining

Investors place importance on ESG information in their investment decisions. Hence, it is needed for technology to extract ESG information from text data. ESG-related information is often contained in integrated reports. However, since the layout of integrated reports differs among companies, it is difficult to create training data for extracting ESG information. In this study, we first extract ESG-related information from securities reports, which can be easily machine-readable. Next, we fine-tune the BERT model using the ESG-related information extracted from the securities reports as training data. Finally, our method extracts ESG-related information from the integrated report by using the BERT model. Our method solves the difficulty of creating training data from integrated reports by using the ESG-related information extracted from securities reports as training data. In evaluation results, our method attains E, S, and G with 93.3%, 91.7%, and 77.4% precision, respectively.