Toward Characterizing Cities with Social Media Images Using Activity Recognition, Topic Modeling and Visualization

Images are at the heart of the digital life, registering what, where, and when people do activities. This rich visual information may allow to characterize cities at spatial granularities unseen with traditional methods. To advance in that direction, here we propose a methodology to explore the activities held in cities, by means of activity recognition (what happens in images), topic modeling (what activities are relevant), and visualization (how to choose the number of topics, interpret them, and where do they appear in the city). We apply this methodology to Santiago, Chile, using images from Flickr, and find promising results for planners and urbanites.