Abstract
The pervasiveness of mobile and sensing technologies today has facilitated the creation of Big Crowdsourced Geotagged Data (BCGD) from individual users in real time and at different locations in the city. Such ubiquitous user-generated data allow us to infer various patterns of human behavior, which helps us understand the interactions between humans and cities. In this article, we aim to analyze BCGD, including mobile consumption check-ins, urban geography data, and human mobility data, to learn a model that can unveil the impact of urban geography and human mobility on the vibrancy of residential communities. Vibrant communities are defined as places that show diverse and frequent consumer activities. To effectively identify such vibrant communities, we propose a supervised data mining system to learn and mimic the unique spatial configuration patterns and social interaction patterns of vibrant communities using urban geography and human mobility data. Specifically, to prepare the benchmark vibrancy scores of communities for training, we first propose a fused scoring method by fusing the frequency and the diversity of consumer activities using mobile check-in data. Besides, we define and extract the features of spatial configuration and social interaction for each community by mining urban geography and human mobility data. In addition, we strategically combine a pairwise ranking objective with a sparsity regularization to learn a predictor of community vibrancy. And we develop an effective solution for the optimization problem. Finally, our experiment is instantiated on BCGD including real estate, point of interests, taxi and bus GPS trajectories, and mobile check-ins in Beijing. The experimental results demonstrate the competitive performances of both the extracted features and the proposed model. Our results suggest that a structurally diverse community usually shows higher social interaction and better business performance, and incompatible land uses may decrease the vibrancy of a community. Our studies demonstrate the potential of how to best make use of BCGD to create local economic matrices and sustain urban vibrancy in a fast, cheap, and meaningful way.
Highlights
Vibrant residential communities are defined as places that show diverse and frequent consumer activities
We propose to analyze and extract the features of spatial configuration from urban geography data and the features of social interaction from human mobility to spot highly vibrant communities, which will be formulated as a ranking-based data mining task
3) Given the obtained features, we develop a novel model to learn the patterns of vibrant communities, by combining pairwise ranking objective and sparsity regularization in a unified probabilistic framework, which is greatly enhanced by simultaneously conducting feature selection and maximizing ranking accuracy
Summary
Vibrant residential communities (vibrant communities for short) are defined as places that show diverse and frequent consumer activities. We propose to analyze and extract the features of spatial configuration from urban geography data and the features of social interaction from human mobility to spot highly vibrant communities, which will be formulated as a ranking-based data mining task next. The spatial configuration features are extracted from the urban geography data including public transportation, road networks, and POIs; the social interaction features are extracted from the human mobility data including bus GPS data, taxi GPS data, and smartphone GPS data. Urban vibrancy promotes the probability that mobile users check into a place, enhances the diversity of urban functions, and improves social interactions and centralization across different categories of outdoor activities. The features of spatial configurations, which can be extracted from urban geography data, such as public transportation, road networks, and POIs. P(yi xi) N (yi fi, σ2) N (yi wuxi, σ2), where N represents the normal distribution
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
