Database Reference
In-Depth Information
Intensity and Duration:
These features quantify the duration and the number
of times that users engage in the system. This set of features includes number of
observations, number of co-location observations, time spent at each location.
Location Diversity:
These features aim to understand the context of all locations.
The features include location frequency and the location entropy. For a location
L
, the location entropy is defined as
Entropy
(
L
)
=−
u
∈
U
P
L
(
u
) log
P
L
(
u
),
where
U
is the set of all users, and
P
L
(
u
) is the probability for a user
u
being at
the location
L
.
Specificity:
These features measure whether two persons meet at locations where
less frequently visited by the public. The tf-idf score penalizes the popular places
that many people frequently visit.
Structural Properties:
These features aim to capture network property of two
users such as mutual neighbors and location overlaps.
The experimental results in [
6
] shows that using a variety of classification methods
such as random forests and support vector machines can achieve precision above
60 % in predicting the online relationships using the mobility features.
5
Aggregate Patterns over Multiple Trajectories
The aggregate patterns describe common paths shared by a set of trajectories or a
cluster of moving objects being spatially close for a long time. In this section, we
first introduce the
trajectories patterns
, which is a concise description of frequent
behaviors in terms of space and time. Then we will present the methods on mining
moving object clusters. Finally, we discuss trajectory clustering methods.
5.1
Frequent Trajectory Pattern Mining
A frequent trajectory pattern is a popular path repeated by many trajectories. Finding
frequent trajectory patterns is helpful in summarizing the historical trajectories and
predicting the future movements. A trajectory pattern [
14
] is used to describe a set of
individual trajectories visiting the same sequence of places with similar travel times.
In trajectory patterns, two notions are important: (1) the geographical locations and
(2) the travel time between locations.
If we assume the locations are already symbolized, frequent sequential pattern
[
1
] can be considered as a simplified trajectory pattern. For example, if many people
go from location
X
,to
Y
and then to
Z
,
X
→
Y
→
Z
will a frequent sequential
pattern. In order to enrich the sequential patterns with
transition time information
between locations, Giannotti et al. [
13
] propose the temporally annotated sequences
(TAS). TAS has the following form:
α
1
−→
s
1
α
2
−→···
α
n
−→
s
n
,
T
=
s
0
