In recent years, integrating ESG (Environmental, Social and Governance) factors into corporate assessment frameworks has become a global priority. However, Vietnam lacks a standardized ESG scoring system, posing a challenge for businesses and investors working towards sustainable development goals. To address this gap, this study focuses on the Vietnamese banking sector - one of the country's most influential sectors - and categorizes the ESG action temporal implementation towards predicting ESG scores based on information extracted from annual reports. As part of this effort, a new dataset of 5816 rows was constructed from 523 annual reports from 37 major Vietnamese banks from 2004 – 2023. Each report is labeled according to pre-defined ESG action temporal, creating a solid foundation for machine learning applications. The study used four machine learning models - SVM, ANN, FCNN and a fine-tuned PhoBERT model. Among these, PhoBERT achieved the highest accuracy, correctly categorizing ESG action temporal with an impressive accuracy of 82%. By constructing an ESG-focused dataset and applying advanced text analytics, this study addresses the lack of an ESG scoring framework in Vietnam and provides a practical approach to integrating ESG considerations into corporate assessment processes. These findings contribute to the advancement of ESG activities in emerging markets and highlight the potential of machine learning in automating ESG assessments.
Bản đồ thống kê
Thống kê nội dung
In recent years, integrating ESG (Environmental, Social and Governance) factors into corporate assessment frameworks has become a global priority. However, Vietnam lacks a standardized ESG scoring system, posing a challenge for businesses and investors working towards sustainable development goals. To address this gap, this study focuses on the Vietnamese banking sector - one of the country's most influential sectors - and categorizes the ESG action temporal implementation towards predicting ESG scores based on information extracted from annual reports. As part of this effort, a new dataset of 5816 rows was constructed from 523 annual reports from 37 major Vietnamese banks from 2004 – 2023. Each report is labeled according to pre-defined ESG action temporal, creating a solid foundation for machine learning applications. The study used four machine learning models - SVM, ANN, FCNN and a fine-tuned PhoBERT model. Among these, PhoBERT achieved the highest accuracy, correctly categorizing ESG action temporal with an impressive accuracy of 82%. By constructing an ESG-focused dataset and applying advanced text analytics, this study addresses the lack of an ESG scoring framework in Vietnam and provides a practical approach to integrating ESG considerations into corporate assessment processes. These findings contribute to the advancement of ESG activities in emerging markets and highlight the potential of machine learning in automating ESG assessments.
