Social media-based emergency management to detect earthquakes and organize civilian volunteers

Abstract

Emergencies occur randomly all over the world. Therefore, emergency management plays a very important role in saving lives in disaster situations. Progress in technology and communication has led to new opportunities in early warning systems and rapid response. Social media has been used to extract useful real-time data regarding disaster situations and individuals’ current conditions. In this paper, we aim to detect earthquakes as quickly as possible by analyzing Twitter data using Support Vector Machine, K-Nearest Neighbors and Naive Bayes algorithms. After detecting disaster event, twitter messages are analysed to distinguish those associated to the detected event and demand requests and volunteer offers are collected. Besides official relief operations, there are also actions that can be taken by civilians voluntarily. Such voluntary actions need optimization to be most effective. Therefore, an optimization process is required following data collection from social media. Collected data must include both demands and voluntary offers, the quantities of supplies and geographical location data. It will then be possible to apply several optimization algorithms and direct volunteers and demanders on the basis of the quickest solution, minimizing the distance they need to travel and maximizing the number of requirements met. The problem can be defined as a classification problem or a facility location problem if predetermined geographical locations are used. Cost can be calculated in terms of distance, initial costs and humanitarian costs, using algorithms such as K-means, mean shift and Density-based spatial clustering of applications with noise, in order to identify exact solutions for facility location problems. Our simulation results show that if supplies are sufficient or assumed to be infinite, each demand can be assigned to the closest facility. But in reality, the method used needs to ensure that demands are met as much as possible. For this reason, a Capacitated Facility Location model with a CPLEX solution proved to be the most efficient method among all the tested algorithms for the selected test data-set in terms of cost and time performance.