Restaurant constitutes an essential part of the tourism industry. In times of uncertainty and transition, restaurant survival prediction is vital for deepening organizations' understanding of business performance and facilitating decisions. By tapping into online reviews, a prevalent form of user-generated content, this study identifies review variance as a leading indicator of restaurants’ survival drawing from data on 2838 restaurants in Boston and their corresponding reviews. Machine learning–based survival analysis shows that models integrating fine-grained review variance (i.e., review rating variance, overall review sentiment variance, and fine-grained review sentiment variance) outperform models that do not account for these factors in restaurant survival prediction before and during the pandemic. Furthermore, in most cases, expert reviews hold stronger predictive power for pre-pandemic restaurant survival than non-expert and all forms of reviews. This study contributes to the literature on business survival prediction and guides industry practitioners in monitoring and enhancing their enterprises.

中文翻译：

---

使用机器学习进行餐厅生存预测：顾客在线评论的方差和来源重要吗？


餐饮业是旅游业的重要组成部分。在不确定和转型时期，餐厅生存预测对于加深组织对业务绩效的理解和促进决策至关重要。通过利用在线评论（用户生成内容的一种普遍形式），本研究根据波士顿 2838 家餐厅及其相应评论的数据，将评论差异确定为餐厅生存的主要指标。基于机器学习的生存分析表明，整合细粒度评论方差（即评论评分方差、总体评论情绪方差和细粒度评论情绪方差）的模型在餐厅生存预测中优于不考虑这些因素的模型。疫情期间。此外，在大多数情况下，专家评论对大流行前餐厅生存的预测能力比非专家和所有形式的评论更强。这项研究为企业生存预测的文献做出了贡献，并指导行业从业者监控和增强其企业。