Machine Learning–Based Classification of ESG News: Environmental Information from Digital Media in Vietnam

Vietnam’s ESG disclosure landscape remains fragmented and report-based ESG data is scarce, leaving stakeholders with limited tools to monitor corporate environmental behaviour in near real time. We address this gap by developing an end-to-end pipeline that classifies Vietnamese ESG news on the environmental (E) pillar and links each classification to the Global Reporting Initiative (GRI) framework through a ChatGPT-based interpretability agent (“ClearESG-E”). The pipeline is benchmarked on a manually labelled corpus of news articles covering firms in the Vietnam Sustainability Index (VNSI) and externally validated on a disjoint set of newly labelled articles. The contribution is threefold: a Vietnamese ESG-news classification framework tailored to a low-resource, lexicon-driven setting; an externally validated evidence of generalisation that the top classifiers retain strong discriminative performance on unseen content; and an interpretability layer that maps each environmentally relevant article to GRI sub-criteria with traceable textual evidence, supporting transparent ESG monitoring in emerging markets.