Harnessing Social Media Data for Text Classification During Emergency

Social media platforms offer a plethora of valuable resources in times of disaster. Social media platforms like Twitter allow the sharing of information by individuals in the time of crisis. In this study, highlights difficulty in differentiating between situation re- lated tweets and are intend to investigate the belongingness of tweets that are related to the disaster or not related to the disas- ter. This work proposes an exploration of n-gram-based feature extraction in the form of unigram, bigram, and trigram. Further combinations of unigram, bigram and trigram have been taken with cut-off value 50%. To intensify performance, the two-step ensemble stacking method is implemented to categorize tweets by enhancing the diversity among the base learners and utilizing the meta-learner to merge the predictions made by base learners. To verify the approach, experimentation done with ensemble method and then compared with linear and non-linear machine learning algorithms. The obtained result shows stacking ensemble method with decision tree as meta learner gives the best results. The findings of the study revealed that our proposed method demonstrates a robust solution to help identify disaster related tweets during a disaster and empower individuals, groups, and organizations to improve communication during emergencies.