IS2Fun: Identification of Subway Station Functions Using Massive Urban Data

Urbanization and modernization accelerate the evolution of urban morphology with the formation of different functional regions. To develop a smart city, how to efficiently identify the functional regions is crucial for future urban planning. Differed from the existing works, we mainly focus on how to identify the latent functions of subway stations. In this paper, we propose a semantic framework (IS2Fun) to identify spatio-temporal functions of stations in a city. We apply the semantic model Doc2vec to mine the semantic distribution of subway stations based on human mobility patterns and points of interest (POIs), which sense the dynamic (people's social activities) and static characteristics (POI categories) of each station. We examine the correlation between mobility patterns of commuters and travellers and the spatio-temporal functions of stations. In addition, we develop the POI feature vectors to jointly explore the functions of stations from a perspective of static geographic location. Subsequently, we leverage affinity propagation algorithm to cluster all the stations into ten functional clusters and obtain the latent spatio-temporal functions. We conduct extensive experiments based on the massive urban data, including subway smart card transaction data and POIs to verify that the proposed framework IS2Fun outperforms existing benchmark methods in terms of identifying the functions of subway stations.