As a new part of public transportation system, urban bicycle-sharing system (BSS) consists of bike stations with various kinds of functions, which have significantly important impact on station planning, user… Click to show full abstract
As a new part of public transportation system, urban bicycle-sharing system (BSS) consists of bike stations with various kinds of functions, which have significantly important impact on station planning, user pricing, advertisement distribution, and so on. After being adopted and deployed in more and more cities, the BSSs accumulate increasingly huge usage data (i.e., trip records) which closely relates to the social and economic activities of users in the city. Therefore, it is possible to take advantage of BSS usage data to infer the functions each station has and then get the functions of regions where the station located. Based on the historical trip records dataset of users, a machine learning algorithm, latent Dirichlet allocation, is adopted to learn the functions of bike stations. Furthermore, $k$ -means clustering algorithm is used to cluster these stations based on their functional profiles. We implement our method using the real-world dataset generated by more than 330 bike stations during three months in capital bikeshare system. The proposed station function discovery method is validated by the analysis of spatiotemporal characteristics on traffic patterns for station clusters and evaluated by the comparison of clustering results with the data of point of interests and station names.
               
Click one of the above tabs to view related content.