Cloud-Based RoBERTa NLP for Assessing Academic Institutions’ Contributions to Sustainable Development Goals

Assessing how Higher Education Institutions (HEIs) align their reports with the Sustainable Development Goals (SDGs) is challenging due to inconsistencies in reporting and the risk of SDG-washing. This study introduces a novel cloud-based SDG classification tool that integrates a RoBERTa-based NLP model with TF-IDF vectorization and LexDG, a localized lexicon designed to enhance domain-specific term recognition. By incorporating regional linguistic variations, the system improves classification accuracy and contextual understanding. The cloud deployment ensures scalability, allowing for seamless institutional adoption. Performance evaluation demonstrates the tool’s effectiveness, achieving $96.77 \%$ accuracy, with a micro-average precision of 0.80, recall of 0.70, and an F1-score of 0.74. These results validate its reliability in identifying SDG-related content, providing HEIs with a data-driven approach for sustainability reporting. Future work will explore advanced deep learning techniques and extend its application to corporate and government sustainability assessments.