Identifying spatiotemporal urban activities through linguistic signatures.

Identifying the activities that individuals conduct in a city is key to understanding urban dynamics. It is difficult, however, to identify different human activities on a large scale without incurring significant costs. This study focuses on modeling the spatiotemporal patterns of different activity types within cities by employing user-contributed, geosocial content as a proxy for human activities. In this work, we use linguistic topic modeling to analyze georeferenced twitter data in order to differentiate different activity types. We then examine the spatial and temporal patterns of the derived activity types in three U.S. cities: Baltimore, MD., Washington, D.C., and New York City, NY. The linguistic patterns reflect the spatiotemporal context of the places where the social media content is posted. We further construct a method to link what people post online to the activities conducted within a city. We then use these derived activities to profile the characteristics of neighborhoods in the three cities, and apply the activity signatures to discover similar neighborhoods both within and between the cities. This approach represents a novel activity-based method for assessing similarity between neighborhoods.