Figure 1
From: Identifying latent activity behaviors and lifestyles using mobility data to describe urban dynamics
Detecting latent behaviors (A) Using individuals’ trajectories, we identify the visits to the different places and the categories of those places. (B) Each individual is described by a M-dimensional (normalized) vector that contains the fraction of visits to each of the 286 categories (visitation pattern) plus the fraction of visits at different times during the day and the week (temporal pattern). (C) Non-negative matrix factorization is used to decompose the matrix of the M-dimensional vectors for each of our N users into a matrix of k different latent behaviors and the corresponding behavior’s weights for each user. Icons designed by bqlqn/flaticon.com and Boston maps produced using Open Street Map data