Publication

Abstract : Societies are significantly impacted by disasters; fast and accurate information is needed in order to respond. Despite challenge like unstructured data and fake news, Twitter can be used in sharing real-time updates. This paper focuses on a robust framework for categorizing disaster-related tweets using dense embeddings such as GloVe, BERT, and FastText, along with machine learning models like Support Vector Machine, Decision Tree, XGBoost, Gradient Boost, Light GBM, Multi-layer Perceptron. The GloVe embeddings with SVM model outperformed all of the other models with an accuracy of 90.09%. LIME and SHAP analyses are implemented to evaluate the significance of each word in a given tweet in order to provide insights into their contributions to model's prediction.