A BERT-Based Model to Analyse Disaster’s Data for Efficient Resource Management

In times of emergency, such as natural disasters, emergencies, pandemics, etc., the victims require rapid aid and resources. The response must be timely and effective, followed by swift recovery management. In recent years, the widespread use of online social media platforms has been phenomenal and has shaped the world around us at an unprecedented rate. We propose to build a model to provide a real-time disaster management strategy that collects data about the natural disaster from social media, verifies it, and classifies it for the need or availability of resources. Furthermore, make the information available to organizations that supply the necessary goods and services. The study uses social media data for two categorization tasks: one for classification of tweets connected to disaster and other for needs assessments. Primarily, tweets fetched from the Twitter (X) streaming API undergo a number of preprocessing stages. Term frequency and inverse document frequency (TF-idf), is then used to convert the data into vector matrices that are needed for the training model. Several machine learning techniques are then deployed for classifying disaster-related tweets. Afterwards, the tweets associated with positive class are further processed, analyzed, and classified for the need or availability of resources. Additionally, to improve the accuracy of the classification, we proposes a BERT (Bi-directional Encoded Representations for Transformers) based Tweets classification and Need Analysis models. The proposed BERT-based model is experimentally evaluated by comparing the prediction outcomes with various machine learning models. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.