Mining Regional Mobility Patterns for Urban Dynamic Analytics

City management plays an important role in the era of urbanization. Understanding city regions and urban mobility patterns are two vital aspects of city management. Numerous studies have been conducted on these two aspects respectively. However, few work has considered combining city region partition and mobility pattern mining together while these two problems are closely related. In this paper, we propose region-aware mobility pattern mining framework, which jointly finds the precise origin and destination region partitions while extracting mobility patterns. We formulate it as an optimization problem of maximizing OD's correlations with spatial constraints. Kernelized ACE, is proposed to solve the problem by learning feature representations that guarantee both objectives. Evaluation results using Beijing's taxi data show that the extracted features are appropriate for this problem and our approach outperforms all the other methods with similar to 0.3% spatial overlap and 86.43% OD correlation. Our case studies on New York City's urban dynamics and Beijing's three-year consecutive analysis also yield insightful findings that reveal city-scale mobility patterns and propose potential improvement for city management.