Recent studies on Airbnb have examined the predictors of room prices, successful reservations and customer satisfaction. However, a preliminary investigation of the listings from twenty-two cities across four continents revealed that a significant number of Airbnb homes remained non-booked. Thus, Poisson count-regression techniques cannot efficaciously explain the effects of predictors of successful Airbnb bookings. To address this gap, we proposed a text mining framework using Hurdle-based Poisson and Negative Binomial regressions. We found that the superhost status, host response time, and communication with guests emerged as the most significant predictors irrespective of geographies. We also found that the instant booking option strongly influences the bookings across cities with incoming business visitors. Additionally, we presented a machine learning-based variable-importance scheme, which helps determine the top predictors of successful bookings, to design customized recommendations for attracting more guests and unique advertisement content on P2P accommodation platforms.

最近对 Airbnb 的研究检查了房价、成功预订和客户满意度的预测因素。然而，对来自四大洲 22 个城市的房源的初步调查显示，仍有大量 Airbnb 房屋未被预订。因此，泊松计数回归技术无法有效解释成功 Airbnb 预订的预测因素的影响。为了解决这个差距，我们提出了一个使用基于障碍的泊松和负二项式回归的文本挖掘框架。我们发现，无论地理位置如何，超赞房东状态、房东响应时间和与客人的沟通都是最重要的预测因素。我们还发现即时预订选项极大地影响了来自商务访客的跨城市预订。此外，我们提出了一种基于机器学习的变量重要性方案，该方案有助于确定成功预订的最高预测因素，以设计定制推荐以吸引更多客人和 P2P 住宿平台上的独特广告内容。