HKEX and PAI Tech collaborated to apply a combination of natural language processing (NLP) and deep learning techniques, developing a comprehensive ability to read, understand and interpret the elements of an annual report. This is collectively referred to as Document Intelligence, and requires processing large sets of training data.
The resulting system downloads annual reports and supplementary announcements, locates disclosures within the content corresponding to each Listing Rule, and then deduces whether issuers are compliant. It then retains the assessment and compliance analysis.
The resulting platform has boosted the breadth and efficiency of annual report assessment, and in 2020 it became part of HKEX’s regulatory toolkit. Through AI-suggested disclosure location and compliance assessment, HKEX can cover review of all required disclosure in the annual reports.
The platform’s performance has steadily increased during the development of the model. Overall efficiency rates for location of annual report disclosures and issuer compliance recommendations for the training set reached 90% and 92% respectively. HKEX then tested the model against 50 previously unseen reports and the efficiency rates were 84% and 85% respectively. It is hoped to continuously improve the efficiency through regular review of data generated from user verification. Future areas of research may include extending the platform to results announcements and other types of regular corporate communications.