Friendship and mobility user movement in location based social networks

Friendship and Mobility: User Movement in
Location-Based Social Networks
Fredrick Awuor
17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2011)
Eunjoon Cho, Seth A. Myers & Jure Leskovec
Stanford University

Presentation Overview
 Introduction
 Characteristics of Check-ins
 Friendship and Mobility
 Model of Human Mobility
 Experimental Evaluation
 Summary
6/10/2015
2

Introduction
 People’s movement and mobility patterns have a higher degree of freedom and variation
 At a global scale human mobility exhibits structural patterns subject to geographic and
social constraints.
 People exhibit strong periodic behavior in their movement as they move back and forth
between their homes and workplaces
 Mobility is also constrained geographically by the distance one can travel within a day
 Mobility may also be shaped by our social relationships
 We are more likely to visit places that our friends and people similar to us visited in the past.
 These hypotheses need answers – which have remain largely unknown due to the fact that
reliable large scale human mobility data has been hard to obtain.
Solution: LBSN
6/10/2015
4

Introduction
 Recently, location based online social networking where users share their current location by
checking in on websites such as Foursquare, Facebook, Gowalla
 Traditional records of calls made by cell phones have been used to track location of cell
phone towers associated with calls
Why is LBSN providing a new dimension in understanding human mobility?
 Cell phones provides coarse location accuracy, LBSN provide location specific data
 Eg. One can distinguish check-in to office in 2nd floor from check-in to a coffee shop in 1st floor of same
building
 Check in to LBSN are usually sporadic while cell phone data provides better temporal resolution
as a user “checks-in” whenever she makes or receives a call
 Both types of data contains network information
 LBSN maintain explicit friendship network while in mobile phones the network can be inferred from the
communication network
6/10/2015
5

Introduction
 Combining data from both mobile phone and LBSN richly make is to answer the hypothesis:
 Geographical Movement (where do we move?)
 Temporal Dynamics (how often do we move?) and
 Social network (how do social ties interact with movement?)
Why is understanding human mobility so important?
 Knowledge of users’ locations can help improve large scale systems
 Such as cloud computing, content-based delivery networks and location-based recommendations
 Accurate models of human mobility are essential for urban planning
 Understanding human migration patterns, and spread of diseases.
 Goal: Analyze the role of geography and daily routine on human mobility patterns as well as the
effect of social ties, i.e., friends that one travels to meet.
 Definition:
 “Check-in” Refer to an event when the time and the location of a particular user is recorded.
6/10/2015
6

Characteristics of Check-ins
Dataset Description
 2 LBSN: Gowalla and Brightkite (some pictures)
 Gowalla: Data collection: Feb 09 – Oct ’10, 6.4M check-ins, 196,591 nodes, 950,327 edges
 Brightkite: Data collection April 08 to Oct ’10, 4.5M check-ins, 58,228 nodes, 214, 078 edges
 Undirected graphs (friendships are undirected)
 Dataset of cell phone location trace data (from a major cell phone service provide in Europe)
 2M users and 450M phone calls over a 455 days
 For each call, nearest cell phone tower of caller and receiver was recorded.
 900M check-ins with spatial accuracy of about 3km
 Social network ties is created between pairs of people that called each other at least 5 times (10 calls
total)
 Network of a 2M nodes and 4.5M edges
6/10/2015
7

Task 1: Check-in behavior of users: spatial and social characteristics of user check-ins
 How far from their homes do people tend to travel?
 How likely are they to meet social network friends at locations that they travel to?
6/10/2015
8
 Distributions follow a power law with
exponential cutoff in which there is an kink
at around 100km.
 Distributions are extremely similar for all
datasets.
 While Brightkite and Gowalla include
check-ins from the whole world, cell phone
data drops off quicker due to the small size
of the country.

Task 2: Distribution of distances between the homes of friends (Kink occurs at around 100Km)
 Shows that probability of 2 friends living a certain distance always decreases quickly at first
but then slows down after the distance between the homes increases above 100Km
 Explanation: Users are geographically non-uniformly spread over the Earth and that humans
cluster in cities.
 Suggests that around 100km is the typical human radius of “reach” as it takes about 1 to 2
hours to drive such distance.
6/10/2015
9

Friendship and Mobility
Focus: Interaction of person’s social network structure and their mobility
Task 1: Moving close to a friend’s home: How likely is person A to travel close to the home of her
friend B (How the location of A’s friend B affects movement of A)
 What we expect:
 People are more likely to move to a place in which they have friends, and that this likelihood
decreases as the distance of travel increases.
 So far we saw that most of our friends live geographically close to us, and thus we would expect
that they impact our movement the most.
 However, as we will see later, this is not the case.
6/10/2015
10

 The probability of visiting a friend’s home levels to
0.3 after 100Km mark
 If a user travels more than 100Km from her home,
there’s 30% chance that they will jump close to an
existing friend’s home
 Probability of visiting a friend’s home remains
constant after 100km mark.
 The # of possible locations one can visit increase with
distance, and the # of friends decreases with distance
as well
 This suggests that the probability of visiting a friend
would decrease with the distance traveled.
 More possible locations to visit and less friends, and
thus smaller probability of visiting a friend
 If people traveled independent of the network
structure then the farther away they move from
home the less likely they are to visit a friend’s home.
6/10/2015
11

 Another observation: the actual influence of a friend on a user making a long distance jump
increases with the distance.
 For example, the relative influence of a friend who lives 1,000km away is 10 times greater than the
influence of a friend who lives 40km away.
Task 2: Influence of friends on an individual’s mobility
To distinguish between an existing friendship causing a user to move to a certain location and a
movement to a certain place that then causes a formation of a new friendship
 Observation:
 On average there is a 61% probability that a user will visit a home of an existing friend.
 However, the probability that a check-in will lead to a new formation of a new friendship is 24%.
 Basically, the influence of friendship on individual’s mobility is about 2.5 times greater than the influence of
mobility on creating friendships.
6/10/2015
12

Task 3: Moving to where a friend has checked-in before:
 Observation 1: The farther away a user travels the more likely that movement is influenced
by a friend.
 The amount that friendship influences movement when traveling long distances (longer than
1,000km) is an order of magnitude higher than the influence when traveling short distances (shorter
than 25km).
 Observation 2: For movement farther than 100km from home the probability of checking-in
at the exact same location as a friend has checked-in in the past remains constant at
around 10%.
Limits of using friendship for predicting mobility.
 Though friends has influence on one’s mobility, only 9.6% of all check-ins in Gowalla and
4.1% of all check-ins in Brightkite were first visited by a friend and then by the user.
 So what’s the limits of using friendship for predicting human mobility?
6/10/2015
13

 In general only a small fraction of users have a high overlap in check-ins with their friends
 Observed strong correspondence between the trajectory similarity of users and probability
of friendship.
 So explore the connection between trajectory similarly of a pair of users and the probability
that they are connected in the social network.
6/10/2015
14
Obsvervation 3: When a pair of users has more than
40% of check-ins in common then the friendship
probability is above 0.3.
 This is a strong presence of social and geographical
homophily

6/10/2015
15
Observation 4: The majority of the users have zero check-ins that were previously checked-in by a
friend
 In Gowalla, 84% of users have less than 20%
of their check-ins that were previously visited
by a friend, and 52% of the users have zero
check-ins that were previously visited by a
friend.
This means that for about 50% of the users, there
is basically no information about their mobility
that could be inferred from their social network.

Task 4: Temporal and geographical periodicity of human movement
 So far we have found that social network influences long distance travel more than short distance
travel, while on the other hand
 We also observed that a relatively small fraction of user check-ins were previously checked-in by a
friend.
What are the non social factors of human mobility?
 Intuitively, we expect locations such as home and work are visited regularly and often during the
same time of the day
 Observation 1: 53% of all check-ins in Brightkite (31% in Gowalla) have been previously
visited by the same user.
This means that if a user checks-in into a place for the first time, there is 53% chance she might return
and check-in again.
 The effect of the social network is about 5 times smaller, i.e., there is only a 10% chance that
a user will check-in to a place where a friend has checked-in before.
6/10/2015
16

Task 5: Explore the connection between geographic and temporal periodicity
 For all days, the early morning hours have the lowest location entropy (i.e., most people are
at home).
 Location entropy increases when people are commuting during the rush hour and in the
evening when they might be out socializing.
 During the work week entropy is lower compared to that of the weekend
 People are commuting to and from work at roughly the same time during the work week, as
opposed to the weekend when peoples’ travel and schedules are less predictable.
6/10/2015
17

Model of Human Mobility
Task: Develop a model of human mobility that can accurately predict future movements of an
individual
 Results so far show strong evidence of geographic (returning to the same places) and
temporal (traveling at regular times of the day) periodicity, and also the increasing relative
effect of the social network structure on an individual’s mobility.
 Formulate a model that incorporates the three ingredients of human mobility: temporally and
geographically periodic movement with the social network structure
1. Periodic Mobility Model (PMM) – built on intuition that majority of human movement is based
on periodic movement between a small set of latent states (location)
 Depending on the time of the day, an individual’s movements will either be centered around home,
work, or somewhere in between the two locations as they “commute” in between them.
2.. Periodic & Social Mobility Model (PSMM) - Extend the PMM with social network influenced
check-in) as a driver for human mobility.
6/10/2015
18

Experimental Evaluation
6/10/2015
19

Summary
 Though human mobility patterns have a high degree of freedom and variation, they also
exhibit structural patterns due to geographic and social constraints.
 Establish what governs human motion and dynamics (Dataset: cell phone & 2 LBSN)
 Observations:
 Humans experience a combination of periodic movement that is geographically limited and
seemingly random jumps correlated with their social networks.
 Short-ranged travel is periodic both spatially and temporally and not effected by the social
network structure, while long-distance travel is more influenced by social network ties.
 Social relationships can explain about 10% to 30% of all human movement, while periodic behavior
explains 50% to 70%.
 Based on findings, developed human mobility model that combines periodic short range
movements with travel due to the social network structure
 The model performs better in predicting the locations and dynamics of future human
movement
6/10/2015
20

Friendship and mobility user movement in location based social networks

Recommended

More Related Content

What's hot (14)

Viewers also liked (20)

Similar to Friendship and mobility user movement in location based social networks (20)

Recently uploaded (20)

Friendship and mobility user movement in location based social networks

Editor's Notes