Finding by Rensselaer Researchers Could Impact
Emergency Management, Infrastructure Development, and Disease
Control
![]()
The closer you live to another person, |
The closer you live to another person, the more likely you
are to be friends with them despite the growing use and impact
of social media, according to a study that drew on data from
the location-based social network provider Gowalla. The study,
by researchers within the Social Cognitive Network
Academic Research Center (SCNARC) at Rensselaer, also
showed that people tend to move in groups of friends, and that
two people chosen at random at a specific event (like a concert
or at a particular store) are unlikely to be friends.
While the findings are seemingly common-sense, the study –
and continued research on social networks – holds a powerful
message for a broad range of applications that rely on accurate
predictions of how people move, such as emergency planning,
infrastructure development, communications networks, and
disease control.
“The ramifications are extremely important because if we
assume that people are moving randomly, we are wrong, and
therefore we will not be prepared for what people actually do,”
said Boleslaw Szymanski, director of SCNARC and the Claire and
Roland Schmitt Distinguished Professor of Computer Science at
Rensselaer. “Where you live really matters: Most of your
friends are concentrated in the place where you live, and as
the distance increases, this concentration rapidly drops.”
The findings also indicate that, even in the digital age,
humans still form friendships based on personal interactions,
said Tommy Nguyen, a Rensselaer graduate student and member of
SCNARC.
“Even though, thanks to the Internet, you can be friends
with anyone on the planet, the likelihood that a person will be
friends with someone in a distant location chosen at random is
far lower than the likelihood that this person will be friends
with someone who lives in close proximity,” said Nguyen.
“Proximity creates a strong boundary for who will be your
friends.”
The study, titled “Using
Location-Based Social Networks to Validate Human Mobility and
Relationships Models,” was awarded the best paper award at
the Second Workshop on Social Network Analysis in Applications
held earlier this year in Istanbul, Turkey. The work continues
the group’s recent investigations into social networks. Last
year, the group published a
landmark paper showing that when just 10 percent of the
population is committed to an opinion without comparable
committed opposition, their opinion will quickly be adopted by
the majority of the society.
The current study drew on the public profiles (friends and
check-ins) of 391,223 users of Gowalla collected between
mid-September and late-October of 2011. Gowalla (which has
since been purchased by Facebook, and is no longer available)
allowed its users to share their geographic location with their
friends through their smart phones in a process known as
“checking in.” The users accumulated a total of around 26
million “check-ins” and 8 million friendship links. Data was
provided to researchers without individual identifications to
protect the privacy of users.
“When a detective wants to solve a crime, they use clues to
draw the big picture,” Nguyen said. “Gowalla provided the
discrete location of the movements of hundreds of thousands of
people – those are clues.”
The data immediately revealed that the likelihood of
friendship between two people decreases as distance increases.
Researchers found that 80 percent of friends of a particular
person live within 600 miles of that person’s home.
“You may have a few distant friends who are holdovers from a
time when you lived elsewhere, or who share a common trait like
family connections or a particular activity, but in general,
the likelihood of friendship decreases as distance increases,”
Szymanski said. “That tells us an important thing which our
findings highlight: Friendship requires constant interactions,
maybe physical presence (making proximity important) because we
prefer to rely on verbal and body language to invoke feelings
of trust in people. That’s very important in friendship.”
The researchers also found that friends tend to move
together.
“If we see two people traveling together, we know first of
all that social relations very much dictate our itinerary when
moving over time. We cannot assume that people move randomly,”
Szymanski said.
Then, the researchers used the data to inform a data-driven
mathematical model predicting the movement of people. Starting
with a model that predicted the movements of individuals at
random, then they refined the model based on the premise that
each individual had friends who lived nearby, and that each
individual frequently moved with their friends.
The “friends” model produced a dramatically different
pattern of movement, and one far more consistent with the data
they studied – data which tracked the actual movements of
Gowalla users.
“If you look at the frequency of movement among people, the
area, and distance they traveled, the model we have developed
reasonably describes their movement,” Nguyen said.
The “friends” model can be used in emergency management,
development of infrastructure, disease control, and can also
help in building friendlier communities, with initiatives like
bicycle sharing, or planning the location of recreational
facilities.
“People travel together, so knowing their social groups
enables us to predict where they move,” Szymanski said. “In
other words, our infrastructure should reflect our social ties
because then it would be aligned with the movements people
would make. That is a helpful insight.”
The researchers are already planning their next project,
which looks at whether the pattern of movement endures through
friends of friends, or further down the chain – what might be
called the “transmissibility” of friendship. Early results seem
to indicate that by the third generation (friend of friend of
friend) all structure has broken down.
“In other words, friend of friend has some value in terms of
movement,” said Szymanski. “But friend of friend of friend
shows not more pattern than a random person.”
Szymanski said the study is also an early indication of the
power that social networks hold not just for users, but for
researchers.
“Gowalla was this super memory of the 100,000 people. Of
course, the users checked in because they hope to meet their
friends. But for us, social media has created new tools to
observe social interactions. It’s unobtrusive, but it is so
powerful that I can’t imagine how we could replicate this study
without this social media tool,” Szymanski said. “For us, it is
invaluable.”