Discovering urban functional regions based on sematic mining from spatiotemporal data

Abstract:

To tackle the problem that the current urban functional regions division are manual completed and do not fully use the spatiotemporal data in urban regions, an approach for detecting urban functional regions is proposed based on sematic mining from spatiotemporal data. In which, a rectangular area of the city is first selected as a research sample and divided it into some valid basis region units according to its buildings. Dirichlet multinomial regression (DMR) topic model is then implemented for the checkin and POI(points of interest) data from Sina weibo in these basis region units and the functional vectors of the basis region units are obtained. Finally，the functional regions are discovered with vector clustering algorithm and POI’s category proportion. The experimental results show that this approach has higher accuracy compared with the kmeans clustering method based on POI density and urban functional area detecting approach based on latent Dirichlet allocation (LDA) topic model. Therefore，The activity patterns of people identified by location checkin data can reveal the differences between urban functional areas and have a good effect on urban geospatial analysis.