Sentimen Analysis Social Media for Disaster using Naïve Bayes and IndoBERT

Abstract

The rapid advancement of information and communication technology has resulted in a significant surge in data, especially text data from social media platforms. This paper presents a sentiment analysis approach using IndoBERT and Naïve Bayes algorithms to classify sentiment related to natural disasters, specifically from a dataset of tweets derived from social media platform X. The focus of this research is to categorize tweets as positive and negative sentiment to provide useful insights in improving disaster response and management, with a focus on tweets related to earthquakes, floods, and the eruption of Mount Merapi. The goal is to assist the government in allocating aid more efficiently and understanding public sentiment during disasters. The methodology used includes data collection, data preparation, labeling, categorization, word weighting using tf-idf, data separation, and classification using Naïve Bayes and IndoBERT algorithms. The results showed that IndoBERT achieved 91% accuracy, while Naïve Bayes achieved 74% accuracy. The study highlights the potential of sentiment analysis in improving disaster preparedness and more effective response strategies.