Database Reference
In-Depth Information
300
a non−friend pair
a friend pair
200
100
0
Mon
Tue
Wed
Thurs
Fri
Sat
Sun
days of the week
Fig. 12.7
Meeting frequency for a friend and a non-friend pair in Reality Mining dataset [
23
]
Category
Variables
Description
Co-location
User mobility
NumObservations
The total number of observations of the user.
√
NumColoc,
NumColocEvening,
Num-
The number of co-location observations of the two users, in total, in the
evening only, and on weekends only.
√
ColocWeekend
NumLocations, NumLocationsEvening,
NumLocationsWeekend
The number of distinct grid boxes where the user or users were observed, in
total, in the evening only, and on weekends only.
√
√
Intensity
and
Duration
NumHours, NumWeekdays, NumDates
The number of distinct hours of the day, days of the week, and calendar
dates that the two users were observed together.
√
ObservationTimeSpan
The difference in seconds between the last and the first location or co-
location observation.
√
√
BoundingBoxArea
The
area
of
the
minimal
axis
aligned
rectangle
that
contains
the
√
√
locations/co-location observations of the user/users.
AvgEntropy,
MedEntropy,
VarEntropy,
The
mean/median/variance/min/max
of
the
location
entropy
at
each
√
√
MinEntropy, MaxEntropy
location/co-location observation of the user/users.
AvgFreq,
MedFreq,
VarFreq,
MinFreq,
The mean/median/variance/min/max of the location frequency at each
location/co-location observation of the user/users.
Location
Diversity
√
√
MaxFreq
AvgUserCount, MedUserCount, VarUser-
Count, MinUserCount, MaxUserCount
The mean/median/variance/min/max of the location user count at each
location/co-location observation of the user/users.
√
√
SchEntropyL,
SchEntropyLH,
SchEn-
The schedule entropy of the user with respect to location, location and hour,
location and day of the week, and location and hour and day of the week.
√
tropyLD, SchEntropyLHD
Mobility
Regularity
SchSizeLH, SchSizeLD, SchSizeLHD
The schedule size of the user with respect to location and hour, location and
day of the week, and location and hour and day of the week.
√
AvgTFIDF, MinTFIDF, MaxTFIDF
The mean/minimum/maximum of the location TFIDF at each co-location of
the two users.
√
Specificity
PercentObservationsTogether
The total number of co-locations of the two users divided by the sum of
each users total number of observations.
√
NumMutualNeighbors
The number of people who have been co-located with both users.
√
NeighborhoodOverlap
The number of people who have been co-located with both users divided by
the number of people who have been co-located with either user.
√
Structural
Properties
LocationOverlap
The total number of distinct places visited by both users divided by the total
number of places visited by either users.
√
Fig. 12.8
Names and descriptions of the mobility features used in [
6
]
The co-locating times could be a discriminative feature to indicate the semantic
relationships. Figure
12.7
shows the meeting frequency with respect to different days
of the week for a friend pair and for a non-friend pair in Reality Mining dataset. It
is shown in the figure that the friend pair meets more on the weekends, while the
non-friend pair meets more during the weekdays. Motivated by this observation,
Li et al. [
23
] propose to mine discriminative time intervals to classify whether two
people are friends. The discriminative interval, namely T-Motif, is the time interval
where there is a significant difference in meeting frequency between friend pairs and
non-friend pairs.
To study how interactions in mobility data correlate with friendships on social
networks, Cranshaw et al. [
6
] propose to build a supervised learning framework
using features extracted from mobility data to predict the online relationship. They
use a location sharing application based on user check-ins on Facebook to obtain the
mobility data from 489 users. Using the mobility data, they propose a set of features
as shown in Fig.
12.8
. The features can be divided into four categories:
