A Novel Airbnb Matching Scheme in Shared Economy Using Confidence and Prediction Uncertainty Analysis

As a shared economy platform, Airbnb allows customers to collaborate and guides them to hosts' rooms. Based on the records and ratings, it attaches great significance to infer users' satisfaction with their rooms. Several essential problems arise when evaluating satisfaction and matching. Data confidence and prediction bias influence the inference performance of the satisfaction. When two users stay in one room, their joint satisfaction also deserves particular research because of the roommate effect. In this paper, a matching model is built based on the inferred satisfaction considering confidence and prediction uncertainties. The satisfaction with the confidence uncertainty is modeled using a normalized variance of the Beta distribution. The algorithms for inferring satisfaction with the prediction uncertainties are divided into two parts: a weighted matrix factorization-based algorithm for individuals and a preference similarity-based algorithm for pairs. Two matching algorithms are proposed with constraints. Finally, extensive experiments using real-world data show the effectiveness and accuracy of the proposed method.