ClimatePulse: Sentiment and Emotion Analysis of Public Discourse on Climate Change in Social Media using BERT, NER, Multilabel Classification, and Spatio-Temporal Visualization

Climate change poses major global challenges with wide-ranging impacts on ecosystems, health, and policy. Public responses to climate policies are increasingly voiced on social media, producing large volumes of data that require intelligent analysis. This study introduces ClimatePulse, a unified system for analyzing Indonesian public opinion on climate change by integrating sentiment analysis, fine-grained emotion classification, and spatio-temporal visualization. The system leverages a fine-tuned BERT model trained on a balanced dataset of 3,256 tweets. Results demonstrate that the model achieves an accuracy of 75.57% and a Macro F1-Score of 0.7556, outperforming traditional baselines like SVM and Logistic Regression. Specifically, the model excels in detecting negative sentiment with an F1-Score of 0.8151, capturing critical public dissatisfaction. Beyond sentiment, the system identifies dominant emotions (e.g., sadness, joy, fear) and visualizes geographic trends through an interactive map, providing actionable insights for policymakers. While challenges remain in classifying ambiguous neutral texts (F1-Score 0.7035) and detecting sarcasm, ClimatePulse effectively bridges the gap between unstructured social media data and data-driven decision-making, directly supporting SDG 13: Climate Action.