Detecting needs of people in a crisis using Transformer-based question answering techniques
Esraa Karam, Wedad Hussein, Tarek F. Gharib · 2021
The usage of social media in everyday life has become a need for keeping up with the news, as well as for making queries and requesting assistance. Following a significant event like a disaster, online media correspondence is a significant part of the emergency reaction. Regardless of the sort of disaster, whether a storm, an earthquake, or a man-made disaster like a riot or a terrorist attack. Social media platforms such as Facebook, Twitter, and others have proven to be efficient communication and coordination tools for disaster victims and other groups. This gathered data can be utilized to react to individuals with the necessary requirements relying upon the kind of help or request mentioned. In this paper, we suggest an approach for extracting the needs of affected people during a crisis and providing the appropriate response using question answering techniques based on natural language processing techniques and neural networks. The approach was tested using Twitter data from various types of emergencies with different question answering techniques such as BERT (Bidirectional Encoder Representations from Transformers), DistilBERT, T5 (Text-to-Text Transfer Transformer Model). The results showed that the BERT transformer the proposed approach has a precision of 0.81, a recall of 0.76, and an f-score of 0.78 for providing appropriate guidelines.