Mining informative words from the tweets for detecting the resources during disaster

Abstract

Millions of tweets are posted on twitter during disaster. Many prior studies discussed about detection of situational and non-situational information occurred during disaster. It is difficult task to detect the tweets related to the resources because tweets related to the resources is a subset of the situational information. During disaster the data are unlabeled. It is not possible to predict the data in supervised classifier without labels. Hence, a classifier based on the informative words for detecting the resources is proposed in this work. It is trained with past data and tested with future events. In this work, the Italy earthquake 2016 data-set is used which is provided by SMERP 2017. First day tweets are used for training the classifier and second and third day tweets are used for testing purpose. The proposed features outperforms than Bag-Of-Words (BOW) in both in-domain and cross-domain schemes.