Abstract—Environmental sustainability policy relies on public attitudes to gauge acceptance and effectiveness, but traditional survey methods fall short in capturing emerging sentiment trends. This study applies machine learning and deep learning techniques, such as Bidirectional Encoder Representations from Transformers (BERT) and Long Short-Term Memory (LSTM), to analyze social media discussions about environmental policies. It turns unstructured data into structured sentiment values and evaluates them using an Extended Policy Analytical Framework. This framework includes region-specific analysis, event-related sentiment trends, emotion profiling, and explainable AI (SHAP-Shapley Additive Explanations) for clarity. BERT outperformed other models, achieving 0.97 accuracy, followed by random forest at 0.94. The results reveal notable sentiment changes around key global policy events, such as COP26 in India (+0.14) and U.S. carbon tax proposals (−0.15), along with emotional trends related to specific issues. A comparative regional analysis showed a performance drop of 5% to 7%, indicating regional differences. Overall, the research demonstrates that explainable AI-driven sentiment analysis can provide useful information to improve policy design and communication.