Characterization of Geographic Regions Based on Georeferenced Data from the Social Web

The characterization of specific places or more general geographic regions is essential to a variety of decision-making processes, particularly in the context of problems related with urbanism or demographic studies. In the context of my MSc thesis, I propose new ways of characterizing geographic regions, through the usage of georeferenced information extracted from location-based social networks and from popular Web 2.0 services, such as Twitter, FourSquare or Flickr. The specific methods that I propose in my dissertation characterize geographic regions with basis on information extracted from publicly available georeferenced photos, shared by the users of Flickr, together with auxiliary information available from raster datasets containing geographic information (e.g., elevation or population density) about the desired locations. Data classification techniques are used to estimate the boundaries of vague regions, or to infer geographic characteristics like the land coverage. The classification methods are based on Support Vector Machines, leveraging on multiple Gaussian kernels to increase the estimation accuracy. An extensive set of experiments attests to the effectiveness of the proposed methods.