ABSTRACT

Natural disasters have caused substantial economic losses and numerous casualties. The demand analysis of relief supplies is the premise and basis for efficient relief operations after disasters. With the widespread use of social media, it has become a vital channel for people to report their demand for relief supplies and provides a way to obtain information on disaster areas. Therefore, we present a topic model-based framework and establish a demand dictionary and a gazetteer that aims to identify the spatial distribution of the demand for relief supplies by using social media data. Taking the 2013 Typhoon Haiyan (also called Yolanda) as a case study, we identify the potential topics of tweets with the biterm topic model, screen the tweets related to demands, and obtain the demand and location information from tweets to study the distribution of the relief supplies needs. The results show that, based on the demand dictionary, a gazetteer and the biterm topic model, the effective demand for relief supplies can be extracted from tweets. The proposed framework is feasible for the identification of accurate demand information and its distribution. Further, this framework can be applied to other types of disaster responses and can facilitate relief operations.

摘要

自然灾害造成了巨大的经济损失和大量人员伤亡。救灾物资需求分析是灾后高效救灾行动的前提和基础。随着社交媒体的广泛使用，它已成为人们报告救灾物资需求的重要渠道，并提供了获取灾区信息的途径。因此，我们提出了一个基于主题模型的框架，并建立了一个需求词典和一个地名词典，旨在通过使用社交媒体数据识别救援物资需求的空间分布。以2013年台风海燕（又称约兰达）为案例，利用biterm主题模型识别推文的潜在主题，筛选与需求相关的推文，并从推文中获取需求和位置信息，研究救援物资需求的分布情况。结果表明，基于需求字典、地名词典和biterm主题模型，可以从推文中提取对救济物资的有效需求。所提出的框架对于识别准确的需求信息及其分布是可行的。此外，该框架可以应用于其他类型的灾难响应，并可以促进救援行动。