Cross-Urban Point-of-Interest Recommendation for Non-Natives

This article describes how understanding human mobility behavior is of great significance for predicting a broad range of socioeconomic phenomena in contemporary society. Although many studies have been conducted to uncover behavioral patterns of intra-urban and inter-urban human mobility, a fundamental question remains unanswered: To what degree is human mobility behavior predictable in new cities—a person has never visited before? Location-based social networks with a large volume of check-in records provide an unprecedented opportunity to investigate cross-urban human mobility. The authors' empirical study on millions of records from Foursquare reveals the motives and behavioral patterns of non-natives in 59 cities across the United States. Inspired by the ideology of transfer learning, the authors also propose a machine learning model, which is designed based on the regularities that they found in this study, to predict cross-urban human whereabouts after non-natives move to new cities. The experimental results validate the effectiveness and efficiency of the proposed model, thus allowing us to predict and control activities driven by cross-urban human mobility, such as mobile recommendation, visual (personal) assistant, and epidemic prevention.