A DEEP LEARNING-BASED COMPARATIVE ANALYSIS OF ESG STRATEGIES IN CHINA AND SOUTH KOREA UNDER IFRS COMPLIANCE

Abstract

 The global framework of Environmental, Social, and Governance (ESG) policies is progressing at a rapid pace, generating a growing need for comparative studies across nations. This research conducts a deep learning-based comparative evaluation of the ESG structures of China and South Korea, examined through the potential alignment with International Financial Reporting Standards (IFRS). Within the scope of bibliometric and systematic literature reviews on ESG standardisation and green infrastructure, a hybrid deep learning model is developed. The model integrates RoBERTa for document-level sentiment assessment, TabNet for the interpretation of ESG indicators, and Multi-Task Transformers (MTL) for concurrent classification and predictive operations. RoBERTa is retrained to capture thematic subtleties, sentiment variations, and the hierarchical prioritisation of ESG strategies across policy documents and corporate sustainability disclosures. TabNet facilitates the interpretation of diverse ESG metrics, enabling the analysis of structured datasets. The MTL framework supports the simultaneous categorisation of focal areas and their validation against international standards, including sentiment compliance with regulatory objectives. Experimental outcomes demonstrate F1 scores of 0.91 for topic identification, 0.88 for compliance detection, and 0.90 for accuracy, all of which outperform traditional baselines. The findings reveal that South Korea utilises a more integrated framework, emphasising ecological infrastructure and the involvement of the public sector, whereas China adopts a more centralised, policy-driven model characterised by state control and mandatory ESG reporting. The study contributes a scalable methodology for establishing ESG standards and harmonising policies internationally, offering stakeholders adaptive guidance to support sustainable and cross-national decision-making.