Construction of a Feature Dictionary and Optimization of an Artificial Neural Network for ESG Information Classification: A Case Study in Vietnam

In the context of the global green transition and Vietnam's commitment to achieving Net Zero by 2050, Environmental, Social, and Governance (ESG) disclosures have become crucial. However, corporate ESG reports in Vietnam often lack quantitative data and transparency. Financial news platforms provide abundant, real-time ESG insights, but processing this massive volume manually is unfeasible. This study pioneers the application of an Artificial Neural Network (ANN) combined with Natural Language Processing (NLP) to automatically classify Vietnamese corporate news into E, S, and G pillars,. Using a dataset of 210 scraped articles, we applied TF-IDF for feature extraction and optimized the ANN's hidden layer through grid search,,. The study successfully established a localized core dictionary of 97 Vietnamese ESG keywords. Performance evaluation showed the Environmental (E) model achieved excellent predictive capacity with 90.48% accuracy and an AUC of 1.000 at 5 hidden nodes,. Although the Social and Governance models exhibited instability due to small test sample sizes, the overall framework demonstrates significant potential. This automated approach provides an objective, real-time screening tool, mitigating “greenwashing” practices, assisting investors in directing green capital, and supporting regulatory bodies in promoting sustainable corporate governance