Database Reference
In-Depth Information
government of the territory. We apply social network analysis techniques to
mobility data. Our aim is to reach a better understanding of human mobility
patterns, using a different perspective based not on the interactions of humans
themselves, but rather on the underlying, hidden connections that reside among
different places. To do so, we apply community discovery algorithms to the
network of geographic areas (i.e., where each node represents a cell or region
of movements), with the aim of finding areas that are densely connected by the
visits of different users. A community discovery algorithm takes as input a graph
and determines a partition of its nodes into communities. Thus, to apply such a
method, it is necessary to extract a network model from the mobility data. As
in Section
10.2
, we adopt census sectors to generalize movement description.
In particular, each trajectory is generalized by the sequence of census sectors it
crosses during the movement.
Generalized movements can be described by means of a weighted, directed
graph
G
(
V,E
) as follows. Each census sector is mapped to a vertex
v
∈
V
.A
directed edge (
u, v
)
∈
E
is placed if there exists at least a movement from
u
to
v
,
u, v
∈
V
. The weight
w
of the edge corresponds to the number of movements
from
u
and to
v
. The graph has an edge (
u, v
)
∈
E
if at least one trip has two
consecutive points such that the first is mapped to census sector
u
and the second
to
v
.
Once the mobility network has been extracted, a community discovery algo-
rithm may be applied to discover groups of nodes, and hence sectors, that can be
aggregated. In particular, we adopt here one of the best performing nonoverlap-
ping community discovery algorithms, namely Infomap. Once the communities
have been discovered, it is possible to link the nodes back to the geography and
define the region covered by each community.
The clustering contains eleven clusters, which are shown in Figure
10.14
.The
clusters determined by the Infomap algorithm are rendered with distinct colors:
the census sectors grouped within the same clusters are drawn with the same
color. As a reference for the actual administrative partition, we have plotted the
boundary of each town. It is worth noting how the cohesion of the sectors within
the same city is preserved. In fact, there are very few episodes of sectors of a
city that are scattered among several clusters, and this happens more frequently
for rural regions. The zones belonging to the urban centers maintain a strong
cohesion. This phenomenon is due to a larger proportion of intracity trips rather
than long-range movements: while the main highways are intuitively associated
with very dense movement, the local movement within each city is greater than
the flow registered in the outer road network. In fact, all the clusters are centered
around the big urban regions, which serve as attractors for the surrounding
mobility. In the few cases where a sector is associated to a cluster of a different
city, it happens that the “misclassified” sector is located near the administrative
