From facebook posts to news headlines: using transformer models to predict post-disaster impact on mass media content

Natural disasters leave scenes of devastation in their wake upon arrival. Sometimes, the destruction is so prominent that reaching the area where the disaster happened becomes challenging-rescuing and communicating with the victims and providing relief become nearly impossible. Nevertheless, some people continuously try to post about the situation on social media, like Facebook. Moreover, the Mass Media utilized the latest technology to search for the latest updates on the impacted area. Analyzing the social and mass media to be aware of the situation can help the authorities take necessary steps. In this research, we intend to develop a multilingual (Bangla and English) system that can predict the post-disaster impact by maneuvering the contents from mass and social media. Therefore, we have created a novel multilingual dataset using the disaster news from Newspaper and posts from Facebook and utilized the modern text processing architecture transformers. For Bangla and English news and posts, our proposed mBERT model achieved accuracy of 89.38 and 87.80%, respectively.