César A. Hidalgo, Elisa E. Castañer

Abstract

Neighborhoods populated by amenities—such as restaurants, cafes, and libraries— are considered to be a key property of desirable cities. Yet, despite the global enthusiasm for amenity-rich neighborhoods, little is known about the empirical laws governing the colocation of amenities at the neighborhood scale. Here, we contribute to our understanding of the naturally occurring neighborhood-scale agglomerations of amenities observed in cities by using a dataset summarizing the precise location of millions of amenities. We use this dataset to build the network of co-location of amenities, or Amenity Space, by first introducing a clustering algorithm to identify neighborhoods, and then using the identified neighborhoods to map the probability that two amenities will be co-located in one of them. Finally, we use the Amenity Space to build a recommender system that identifies the amenities that are missing in a neighborhood given its current pattern of specialization. This opens the door for the construction of amenity recommendation algorithms that can be used to evaluate neighborhoods and inform their improvement and development.