Towards a Comparative Science of Cities: Using Mobile Traffic Records in New York, London, and Hong Kong - Computational Approaches for Urban Environments - page 353

15.1 Introduction

As digital technologies are becoming more and more widespread, big data created
by recording the digital traces left behind human activities become a powerful mean
to study various aspects of human behavior. Many of these aspects can be described
with telecommunications data which nowadays become global. The exploration of
these data provides new perspectives, revealing characteristic usages and regular
dynamic patterns at both the individual and collective scale. At the same time,
the increasing urbanization of the world's population is deeply affecting urban
environments, and it is crucial to develop theoretical frameworks as well as real-time
monitoring systems to understand how the individual dynamics shape the structure
of our cities in order to make better planning decisions.
In the past years, several studies have shown that it was possible to use
telecommunication data to get a fresh view at the spatiotemporal dynamics within
a city. In a now-famous paper, Eagle and Pentland ( 2006 ) showed that it was
possible to decompose mobile phone activity patterns of university students into
regular daily routines and that these routines were linked to each student's major
and also to employment levels. Building upon this work, González et al. ( 2008 )
studied the trajectory of 100,000 anonymized mobile phone users to reveal statistical
regularities in human trajectories. This paper, along with other seminal work
(Candia et al. 2008 ; Song et al. 2010 ), has since generated a research field dealing
with human mobility as understood from digital traces (Kang et al. 2013 ).
In parallel, focusing on records aggregated on spatial locations rather than on in-
dividuals, new approaches have been initiated to describe urban landscape based on
mobile phone usage patterns (Jacobs-Crisioni and Koomen 2012 ; Loibl and Peters-
Anders 2012 ; Ratti et al. 2006 ; Reades et al. 2007 , 2009 ; Calabrese et al. 2011 ;Sun
et al. 2011 ), to explore the issue of regional delineation (Amini et al. 2014 ; Kung
et al. 2013 ; Ratti et al. 2010 ; Sobolevsky et al. 2013 ), to estimate population density
(Girardin et al. 2009 ;Kangetal. 2012 ; Rubio et al. 2013 ; Vieira et al. 2010 ), or to
identify social group and social events (Traag et al. 2011 ). In particular, by mea-
suring mobile phone data on a 500m by 500m “pixel” grid in Rome (Italy), Reades
et al. ( 2009 ) especially used a variant of principal component analysis to cluster
these pixels into regions with similar patterns of usage and made a qualitative link
between these patterns and the number of businesses on the corresponding areas.
This last paper is an example of a line of research dealing with the identification
of a specific land use type (Caceres et al. 2012 ; Calabrese et al. 2010 ). Other papers
have focused on methods to build classification of several land use types based either
on (voice calls or SMS) mobile phone patterns (Andrienko et al. 2013 ; Becker et al.
2011 ; Pei et al. 2013 ; Soto and Frías-Martínez 2011 ; Toole et al. 2012 ), taxi trip data
(Liu et al. 2012 ), or Twitter data (Frias-Martinez et al. 2012 ). These studies used dif-
ferent types of methods, from simple clustering to advanced neural network models.
A common feature of these papers is that they are limited to the study of a single
spatial entity (in general a city) that they study through one type of digital data.
This statement raises some questions: is the behavior detected on one type of
mobile phone activity independent of the other type, i.e., is it the same to look at