Spectral clustering for sensing urban land use using Twitter activity

Abstract

Individuals generate vast amounts of geolocated content through the use of mobile social media applications. In this context, Twitter has become an important sensor of the interactions between individuals and their environment. Building on this idea, this paper proposes the use of geolocated tweets as a complementary source of information for urban planning applications, focusing on the characterization of land use. The proposed technique uses unsupervised learning and automatically determines land uses in urban areas by clustering geographical regions with similar tweeting activity patterns. Three case studies are presented and validated for Manhattan (NYC), London (UK) and Madrid (Spain) using Twitter activity and land use information provided by the city planning departments. Results indicate that geolocated tweets can be used as a powerful data source for urban planning applications.