Detection of situational information from Twitter during disaster using deep learning models

Abstract

Twitter is an excellent resource for communicating between the victims and organizations during a disaster. People share opinions, sympathies, situational information, etc., in the form of tweets during a disaster. Detecting the situational tweets is a challenging task, which is very helpful to both humanitarian organizations and victims. There is a chance that both situational and non-situational information may be present in a tweet. Most of the existing works focused on identifying single-information-type tweets like situational information, actionable information, useful information, etc. Detecting the mixture of situational and non-situational information tweets remains a challenging task. Although existing works designed an SVM classifier using low-level lexical and syntactic features for classifying situational and non-situational tweets, their method does not work well for a mixture of situational and non-situational information tweets. This paper addresses the problem of detecting the situational tweets using different deep learning architectures such as Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Bi-directional Long Short-Term Memory (BLSTM) and Bi-directional Long Short-Term Memory with attention (BLSTM attention). Moreover, deep learning models are applied to Hindi language tweets besides English language tweets for identifying the situational information during a disaster. Some of the tweets are posted in the Hindi language, where the information is not available in the English language in countries like India during the disaster. Experiments are performed on various disaster datasets such as Hagupit cyclone, Hyderabad bomb blast, Sandhy shooting, Nepal Earthquake and Harda rail accident in both in-domain and cross-domain. The results of deep learning models demonstrate that it outperforms the existing traditional approach, such as the SVM classifier with low-level lexical and syntactic features for detecting the situational tweets during the disaster. Additionally, to our best knowledge, this is the first attempt in applying the deep learning models to identify the Hindi language situational tweets during the disaster.