ESG (Environmental, Social, and Governance) is a set of standards for evaluating a company’s non-financial performance, becoming crucial in global trade and integration. Despite the rise of ESG evaluation systems, inconsistencies in rating criteria lead to unreliable results. Extracting ESG data from diverse media can offer more objective insights, but automating this classification remains challenging. This paper proposes a model to automatically identify ESG-related information. A dataset was constructed by manually labeling samples, supplemented with translated global ESG content in Vietnamese. The fine-tuned ESG classifier achieved an accuracy of 94.66% in a 4-class classification task covering Environmental-E, Social-S, Governance-G and Irrelevant-I categories. To evaluate the effectiveness of the proposed model, experiments were conducted, showing two key results. First, the model demonstrated high generalization performance, achieving 94.66% accuracy on sectors not included in the training data. Second, it was able to accurately extract ESG-related content from multi-source text data. The training dataset was compiled from reputable sources using manual labeling and further expanded through pseudo-labeling techniques. For future work, the research team plans to extract and classify ESG information from a broader variety of data sources, aiming to further enhance the model’s robustness and classification performance.