Airbnb Tenant Repeat Occupancy Prediction Model Based On Multimodal Deep Learning

With the rapid development of internet technology,sharing economy platforms in various industries have sprung up like mushrooms after rain.These platforms leverage the information sharing mechanism of the internet to efficiently link resources between providers and demanders,greatly improving the utilization rate of idle resources and bringing enormous economic benefits to society.Among many sharing economy platforms,online short-term rental platforms represented by Airbnb and Xiaozhu have had a disruptive impact and influence on the tourism and hotel industries.Through these platforms,guests can enjoy more distinctive and personalized accommodation services at prices far below those of hotels.While helping landlords to earn economic benefits from their idle properties,it also provides convenience and opportunities for cultural exchange within and outside the region.To some extent,shared short-term rentals have already changed people’s travel consumption patterns and perceptions.After more than a decade of rapid development,the growth of individual landlords and property management companies on shared short-term rental platforms worldwide has gradually slowed down.For shared short-term rental platforms that primarily serve existing customers,guest loyalty to the platform has become an important factor in determining whether online short-term rental platforms can develop in a healthy manner.Repeat stays are an important manifestation of customer loyalty,however,there is still a significant lack of research on the repeat stay behavior of guests on shared short-term rental platforms.Existing research mainly focuses on surveys and simulated studies of guests’intentions to make repeat purchases,with a primary focus on the motives and influencing factors for guests’ repeat stays.In addition,existing studies on repeat purchase prediction mainly focus on e-commerce platforms for general products,which have significant differences in characteristics compared to the service experience and resource use characteristics of the sharing economy.The former focuses on the quality of the product itself,while the latter focuses on the quality of service,the richness of the experience,and the establishment of social relationships.Moreover,from a methodological perspective,previous studies mainly focused on product attributes,user behavior characteristics,and primarily used single-modal numerical and enumerated features.In the field of shared short-term rentals,customer demands are more diverse,and factors considered are more complex.The amount of consumption is relatively large,and the uncontrollability of service quality is also higher.Therefore,customers tend to make purchasing decisions based on more comprehensive and multi-modalinformation.To address these issues,this study conducted research on predicting repeat stays by Airbnb customers using multi-modal deep learning.The main research achievements are as follows:1.Integrated multiple sources of multi-modal data for feature extraction of repeat stays.In the process of using online short-term rental platforms to make accommodation decisions,customers not only pay attention to data such as price,geographic location,and ratings but also consider the image of the landlord and the property reflected by textual and image data.To address this phenomenon,this study integrated numerical,textual,and image data into feature engineering,enriched the feature dimensions of repeat stay behavior,and expanded the feature description angle.According to the experimental results of core feature extraction,textual and image category features accounted for about 37%of the core features of repeat stay behavior,and had a significant impact on predicting customer repeat stay behavior.2.Based on deep learning algorithms,this study conducted research on predicting customer repeat stay behavior and established a customer repeat stay prediction model based on the wide-deep deep learning framework.This method can adapt to continuously changing online short-term rental platforms.The SMOTE optimization algorithm and tree model high importance feature selection algorithm were used to address the problems of sample imbalance and feature selection.Compared with other machine learning methods,the accuracy of this research model was the highest,with an F1 score exceeding 0.90.Finally,this study summarized and reflected on its limitations and provided suggestions for online short-term rental platform-related work and landlords.