摘要:Disaster risk management, including response and recovery, are essential elements of sustainable development. With the recent increase in natural hazards, the importance of techniques to understand, model and predict the evacuation and returning behavior of affected individuals is rising. Studies have found that influence from real world social ties affects mobility decisions during disasters. Despite the rapid spread of social media platforms, little has been quantitatively understood about the influence of social ties on online social media on such decisions. Information provided by who at what timing influences users’ decision-making process by how much during disasters? In this study, we answer these research questions by proposing a data-driven framework that can predict post-disaster mobility decisions and simultaneously unravel the influence of various information on online social media. More specifically, our method quantifies the influence of information provided by different types of social media accounts on the peoples’ decisions to return or stay displaced after evacuation. We tested our approach using real world data collected from more than 13 million unique Twitter users during Hurricane Sandy. Experiments verified that we can improve the predictive accuracy of return and displacement behavior, and also quantify the influence of online information. In contrast to popular beliefs, it was found that information posted by the crowd influenced the decisions more than information disseminated by official accounts. Improving our understanding of influence dynamics on online social media could provide policy makers with insights on how to disseminate information on social media more effectively for better disaster response and recovery, which may contribute towards building sustainable urban systems.