Identifying Damage-Related Social Media Data during Hurricane Matthew: A Machine Learning Approach

Previous research used keywords like Hurricane Matthew/Sandy to filter the disaster- and damage-related social media data. However, various Twitter data containing these keywords were not describing the disaster events or their impacts. Meanwhile, machine learning demonstrates its potential for classifying social media data. Nevertheless, very limited existing research employs this approach for identifying damage-related social media data. This paper introduces the machine learning approach for identifying the damage-related social media data. Naïve Bayes, support vector machine (SVM), and decision tree are employed for training the classifier. The 10-folder cross-validation method is utilized for evaluating the performance of these three classifier models. Naïve Bayes model demonstrates the most reliable results. This paper provides a new solution for filtering the damage-related social media data during natural disasters. The manually annotated Twitter data can be used for classifying social media data in future disaster events.