Twitter-based Urban Area Characterization by Non-negative Matrix Factorization

Due to the remarkable growth of various social networks boosted by the pervasive mobile devices, massive crowds can become social sensors which can share microbolgs on a variety of social situations and natural phenomena in urban space in real-time. In order to take advantages of the novel realm of crowd-sourced lifelogs to characterize urban areas, we attempt to explore characteristics of complex and dynamic urban areas by monitoring crowd behavior via location-based social networks. In particular, we define social conditions consisting of crowd's experiential features extracted from the analysis of Twitter-based crowd's lifelogs. Then, we explore latent characteristic faces of urban areas in term of 5-dimensional social conditions by applying Non-negative Matrix Factorization (NMF). In the experiments with massive geo-tagged tweets, we classify urban areas into representative groups based on their latent patterns which enable to comprehensively understand images of the urban areas focusing on crowd's daily lives.