Extracting Urban Functional Regions from Points

Data about points of interest (POI) have been widely used in 7 studying urban land use types and for sensing human behaviors. However, it 8 is difficult to quantify the right mix or the spatial relations among different 9 POI types indicative of specific urban functions. In this research, we develop a 10 statistical framework to help discover semantically meaningful topics and func11 tional regions based on the co-occurrence patterns of POI types. The framework 12 applies the latent Dirichlet allocation (LDA) topic modeling technique and in13 corporates user check-in activities on location-based social networks. Using a 14 large corpus of about 100,000 Foursquare venues and user check-in behaviors 15 in the ten most populated urban areas of the United States, we demonstrate 16 the effectiveness of our proposed methodology by identifying distinctive types of 17 latent topics and further, by extracting urban functional regions using K-means 18 clustering and Delaunay triangulation spatial constraints clustering. We show 19 that a region can support multiple functions but with different probabilities, 20