Geospatial Sentiment Analysis of Negative Comments on the 2024 Election Using the Robustly Optimized BERT Approach (RoBERTa)Geospatial Sentiment Analysis of Negative Comments on the 2024 Election Using the Robustly Optimized BERT Approach (RoBERTa)Geospat

This study develops a geospatial sentiment analysis system to detect and map hate speech related to the 2024 Election using the Robustly Optimized BERT Approach (RoBERTa). The dataset consists of 11,903 social media comments that have undergone comprehensive preprocessing, including text normalization, stopword removal, and stemming. The RoBERTa model was implemented using 10-fold cross-validation for multi-class classification (HS_Weak, HS_Strong, Not_Abusive) and achieved an average accuracy of 91.54% (±1.08%), with a final model accuracy of 94.29%. Geospatial analysis using geocoding and Folium visualization revealed that 75% of the data originated from Indonesia, with the highest concentration in the Jakarta area. The distribution of hate speech showed consistent patterns between Indonesia (45.6% hate speech) and outside Indonesia (44.3% hate speech), with the HS_Strong category dominating at 96.4%. Heatmap analysis identified hate speech hotspots on the island of Java and a global distribution across various continents. The findings confirm the effectiveness of RoBERTa for sentiment analysis in the Indonesian language and provide valuable insights into the geographic patterns of hate speech in the context of digital politics, which can be used to develop mitigation strategies and real-time monitoring systems.