Understanding timely evolutions of public responses during disasters can help crisis response managers design and implement response strategies. Recent studies mainly employed public sentiment and concerns to investigate public responses with social media data. However, these studies neglected social media users' post frequencies for sentiment analysis which can exaggerate the impact of users with high post frequencies. A sentiment baseline is also missing to reveal disaster impacts on public sentiment. Moreover, a quantification index to represent evolutions of public concerns across the disaster periods is necessary but not investigated yet. In order to bridge the above-mentioned research gaps, this research proposes to analyze social media users' post frequencies and employ the annual average sentiment as a sentiment baseline. This paper innovatively employs the LDA (Latent Dirichlet allocation) topic model to calculate the weights and sentiment for topics in public expressions. To validate this method, a case study of Hurricane Matthew is implemented for investigating the evolutions of public responses. The results show public sentiment in most affected regions has received negative impacts. Public expressions mainly concentrate on crisis-related topics and their sentiment towards these topics varies significantly across the whole disaster periods. The findings can help crisis response managers understand the public's special concerns and panics with the evolution of disasters and further support their design and implementation of effective response strategies.

了解灾难期间公共响应的及时演变可以帮助危机响应管理者设计和实施响应策略。最近的研究主要利用公众情绪和关注点来调查社会媒体数据对公众的反应。但是，这些研究忽略了社交媒体用户的发帖频率来进行情感分析，这可能会夸大发帖频率较高的用户的影响。人们也缺乏情绪基线来揭示灾难对公众情绪的影响。此外，有必要采用量化指数来表示整个灾难期间公众关注的变化，但尚未进行调查。为了弥合上述研究差距，本研究建议分析社交媒体用户的发帖频率，并以年度平均情绪作为情绪基线。本文创新地采用了LDA（潜在狄利克雷分配）主题模型来计算公共表达中主题的权重和情感。为了验证该方法，对马修飓风进行了案例研究，以调查公众反应的演变。结果表明，大多数受影响地区的公众情绪受到了负面影响。公开表达主要集中在与危机相关的主题上，在整个灾难期间，他们对这些主题的看法差异很大。这些发现可以帮助危机应对经理了解公众对灾害演变的特殊关注和恐慌，并进一步支持他们设计和实施有效的应对策略。