A Neural-Based Approach for Detecting the Situational Information From Twitter During Disaster

Twitter is widely considered an essential social media used during a disaster. For the past five years, there has been a great surge in the use of Twitter during the disaster. A large amount of useful information is posted on Twitter during an emergency, along with the users' sympathies and opinions. Therefore, a reliable methodology is needed for extracting useful information from the posted tweets during the disaster. This article is focused on identifying the situational tweets during a disaster. A neural-based approach is developed based on the combination of the RoBERTa model and feature-based method for identifying the situational tweets during a disaster. Extensive experiments are performed on various disaster data sets such as Typhoon Hagupit, Hyderabad bomb blast, Sandy Hook shooting, Nepal earthquake, and HarDerail accident disaster data set. The proposed method is compared with various deep learning models such as convolutional neural network (CNN), long short term memory (LSTM), bidirectional long short term memory (BLSTM), and bidirectional long short term memory with attention (BLSTM attention). Experimental results demonstrate that the proposed method outperforms the existing methods on different disaster data sets.