<![CDATA[This study analyzes public sentiment towards the dissemination of earthquake information by BMKG through the X application with modeling using the Support Vector Machine (SVM) and Random Forest (RF) algorithms. Data were collected from 2,711 tweets mentioning @infoBMKG using tweet-harvest, then processed through the stages of case folding, cleaning, tokenization, slang normalization, stopword removal, and stemming. Automatic sentiment labeling was performed using a hybrid approach of VADER and InSet Lexicon. Feature representation used TF-IDF (Term Frequency–Inverse Document Frequency) with 1,000 features and data distribution 80% train and 20% validation. The results show that RF achieved an accuracy of 81.92% and SVM 81.17%, with almost identical Macro F1 (RF: 0.7445; SVM: 0.7443). Neutral sentiment indicates informative tweets without emotional content (61.75%), negative sentiment represents the public's emotional response that is not solely intended as a form of negative assessment of BMKG (24.71%), and positive sentiment is an expression of appreciation, gratitude, and hope for the delivery of information (13.54%). SVM excels in cross-validation stability (std ±0.0574) and negative sentiment recall (0.71), making it more suitable for real-time disaster communication monitoring.]]>
Copyrights © 2026
