Prompt Learning for ESG (Environmental, Social and Governance) report analysis (ARD/279)

Following rapid growth in Environmental, Social and Governance (ESG) investing over the past decade, companies listed on the Stock Exchange of Hong Kong are now required to issue annual ESG reports. While ESG has been recognised as crucial in maintaining Hong Kong’s position as a major international financial centre, there is no universally accepted standard for ESG reporting. ESG reports are produced by different companies ranging from 10 to over 200 pages, and the ESG ratings for the same company also vary widely depending on the different rating providers. These diversified ESG reports, lacking easily comparable ESG metrics and ratings, have created difficulties for different parties including regulators, investors and listed companies. The objective of this project is to develop an AI-enabled ESG Report Analysis (ERA) system to provide assistance to these parties in compiling, analysing and utilising ESG reports. The system automatically collects, analyses and extracts key information from companies’ ESG reports to provide a comprehensive view of the ESG status of the industries they belong to. The ERA system also identifies potential areas for improvement, and outlines ways of doing this. The proposed solution utilizes prompt learning methods to automatically analyse ESG reports and generate quality improvement suggestions. Research shows prompt learning dramatically outperforms the standard fine-tuning procedures in low-resource(few-shot) learning. Hong Kong would strive to achieve carbon neutrality before 2050 and it is our public engagement on a long-term decarbonization strategy by building more efficient AI technology with less carbon footprint such as prompt learning models.