Identifying initiatives that influence the decision-making process of individuals in the aftermath of extreme natural events is a critical task in post-disaster recovery research. Due to the diversity of disaster-induced physical and psychosocial damage, as well as the complexity of human behavior, a comprehensive understanding of contributing factors requires a collective effort. The growth of social media platforms with millions of users provides researchers with an exceptional opportunity to conceptualize spatial patterns and communal behaviors. This longitudinal study proposes a multistep machine learning algorithm to understand such recovery decisions using social media data. Two publicly available databases, New York City tax lot data and 109 million geotagged tweets from the period October 2012–October 2014 were used to explore residents’ recovery decisions in the two years following Hurricane Sandy. The results reveal that communities with more tweets about social interactions and fewer tweets related to infrastructure and assets were more likely to rebuild rather than relocate.

识别在极端自然事件发生后影响个人决策过程的计划是灾难后恢复研究中的关键任务。由于灾害造成的身体和社会心理伤害的多样性，以及人类行为的复杂性，对影响因素的全面理解需要集体的努力。随着数百万用户的社交媒体平台的增长，为研究人员提供了一个特殊的机会来概念化空间模式和公共行为。这项纵向研究提出了一种多步机器学习算法，以了解使用社交媒体数据进行的此类恢复决策。两个公开可用的数据库，使用2012年10月至2014年10月期间的纽约市税收批数据和1.09亿条带有地理标记的推文，探讨了飓风桑迪过后的两年中居民的恢复决定。结果表明，关于社交互动的推文更多，与基础设施和资产相关的推文更少的社区更有可能重建而不是搬迁。