Collaborative Emergency Decision-making: A Framework for Deep Learning with Social Media Data

Emergency decision-making (EDM) problems based on social media data have recently attracted considerable attention. However, few studies have considered collaborative EDM based on public opinion and expert knowledge. To improve the effectiveness and interpretability of EDM, we propose a knowledge+opinion driven multi-phase collaborative emergency decision-making model, which combines social media data that represents public opinion with the knowledge and experience of experts. First, a text-mining algorithm extracts the keywords and their weights from the social media data. Then, we define 2-tuple emergency attributes to simplify and quantify the keywords with social media data. Furthermore, a sentiment analysis model based on the XLNet-Att deep learning algorithm is proposed to obtain sentiment polarities for emergencies and provide timely support for government EDM in the future. Moreover, a real-world case concerning the Southern China flood disaster in 2020 is applied to validate our proposed model. We find that for similar emergencies, the focus of public attention have similar characteristics at different periods, and the analysis results show different perspectives of public attention to emergencies at different stages, providing reliable data and experience support for future EDM of similar emergencies. Finally, we conduct a sensitivity analysis to demonstrate the stability of our deep learning model and a comparative study using existing models to verify the effectiveness of our model.