A New Approach Based on Twitter Sentiment Analysis and Machine Learning in Post-Disaster Relief Management

Abstract

Throughout history, natural and man-made disasters have caused significant destruction in biological, psychological, social, and economic areas, and their complete prevention has not been possible. Disaster management is essential to reduce the devastating effects of events such as earthquakes, fires, landslides, floods, erosion, droughts, and avalanches, whose frequency is increasing. In this process, data sharing and the rapid, coordinated flow of information are crucial for post-disaster recovery. The earthquake in our country on February 6, 2023, once again demonstrated the importance of data sharing in disaster management. After the disaster, social media platforms, especially X, emerged as valuable data sources. This study aims to classify and analyze data obtained from X using machine learning algorithms and sentiment analysis in the context of disaster management. It seeks to understand the emotional responses of society during and after disasters and to support rapid response and relief operations through the effective use of this information. Considering the speed and constantly updated nature of data, various algorithms were compared for accurate and fast classification. In the data pre-processing, natural language processing techniques and sentiment analysis methods were employed to identify the emotional states of society following the disaster. The experimental results demonstrate that Random Forest, Multi-Layer Neural Networks, and GRU models achieve a 99% accuracy rate, thereby exhibiting a substantial improvement over the Naive Bayes (91%) and Extra Decision Tree (94%) approaches. In class-based evaluations, it was observed that GRU and neural network-based methods, in particular, produced high precision and recall values (0.9 and above) across all categories, demonstrating consistent performance despite the imbalanced class distribution. This study highlights the importance of operations research and industrial engineering and demonstrates the potential of data science to enhance disaster response strategies.