Abstract
In emergencies, X (formerly Twitter) has emerged as an essential and fast information-sharing tool. A mechanism based on TF-IDF (Term Frequency-Inverse Document Frequency), WordNet, an ensemble (stacking) learning model, and multilingual support is proposed to classify disaster-related tweets obtained from the X platform. Moreover, this proposed model is evaluated on eight popular classifiers for classifying textual data. An ensemble of MLP, MNB, CNB, and SVC yields a good overall model with 81.36% accuracy and an AUC of 0.80. The proposed model can predict whether a tweet is related to a disaster by using an unlabeled dataset of tweets. It can classify tweets in multiple languages due to its multilingual support. The model uses Turkish, a low-resource language, to classify disaster-related tweets. Current research shows the application of this model in English and Turkish. This research uses two publicly available datasets of disaster-related tweets. The proposed model achieved good performance metrics on both labeled and unlabeled datasets. The model can be utilized by disaster relief authorities to filter out actual disaster-related tweets automatically and efficiently from many other fake and "clickbait tweets."

This publication has 35 references indexed in Scilit: