Enhancing Disaster Tweet Classification with Ensemble Models and Multiple Embeddings

Social media, particularly Twitter, is a vast source of information where users post tweets, leading to a significant volume and variety of shared content. During natural disasters, the high number of disaster-related tweets often makes it a trending topic. However, not all of these tweets provide relevant information about the disaster, as some may merely use disaster keywords without discussing the event. In order to address the obstacle at hand, the present study suggests a model consisting of a combination of FastText, BERT, and Globe embeddings together with an LSTM foundational model. The objective of this ensemble methodology is to encompass a wide range of tweet semantics and contextual details The utilization of the LSTM model serves to effectively capture the sequential dependencies and temporal information present within the tweet data. Performance evaluation utilizes established metrics like Precision, Recall, Accuracy, and F1-Score to comprehensively assess the model’s ability to classify between disaster and non-disaster tweets. The primary objective is to develop a robust supervised learning classifier that accurately identifies genuine disaster-related tweets. By integrating multiple embeddings and LSTM, the proposed model enhances the accuracy and reliability of disaster tweet classification. This research contributes to real-time emergency response and decision-making by effectively utilizing social media information.