Discovering Urban Travel Demands Through Dynamic Zone Correlation in Location-Based Social Networks

Abstract

Location-Based Social Networks (LBSN), which enable mobile users to announce their locations by checking-in to Points-of-Interests (POI), has accumulated a huge amount of user-POI interaction data. Compared to traditional sensor data, check-in data provides the much-needed information about trip purpose, which is critical to motivate human mobility but was not available for travel demand studies. In this paper, we aim to exploit the rich check-in data to model dynamic travel demands in urban areas, which can support a wide variety of mobile business solutions. Specifically, we first profile the functionality of city zones using the categorical density of POIs. Second, we use a Hawkes Process-based State-Space formulation to model the dynamic trip arrival patterns based on check-in arrival patterns. Third, we developed a joint model that integrates Pearson Product-Moment Correlation (PPMC) analysis into zone gravity modeling to perform dynamic Origin-Destination (OD) prediction. Last, we validated our methods using real-world LBSN and transportation data of New York City. The experimental results demonstrate the effectiveness of the proposed method for modeling dynamic urban travel demands. Our method achieves a significant improvement on OD prediction compared to baselines. Code related to this paper is available at: https://github.com/nicholasadam/PKDD2018-dynamic-zone-correlation .