Social Media for Post-Disaster Relief: Mapping Needs and Availabilities to UNOCHA Resource Classes

What is it about?

In the aftermath of a natural disaster, the key objective is to save human life by providing life-saving resources in real-time. However, the major challenge in coordinating relief operations is the lack of real-time information on resource-needs and resource-availabilities in the disaster-affected region. For the last few decades, Online Social Media (OSM) has become an important source of such real-time information during disasters. Specifically, for disaster events that occur in urban regions, due to the ubiquity of smartphones and availability of stable internet service, the affected population is more inclined to post the information regarding resource-needs and resource-availabilities in OSM. Hence, for disaster events that occur in urban regions, we propose to use Online Social Media as a source of such real-time information. In the present study, we specifically discuss the challenge of mapping social media posts (microblogs) to resource classes as per UNOCHA guidelines. Subsequently, we have attempted to automate the class-wise resource segregation of the tweets using a multi-label classification approach. To this end, we have experimented with several traditional and neural network based classifiers. We have also utilised the transfer learning approach through BERT pre-trained model. We have found the BERT pre-trained model has outperformed all traditional as well as neural network based classifiers in terms of F-Score.

Featured Image