Presentation slides at KDE seminar 2013/02/18, which introduces the paper "The Length of Bridge Ties: Structural and Geographic Properties of Online Social Interactions."
1. The
Length
of
Bridge
Ties:
Structural
and
Geographic
Proper9es
of
Online
Social
Interac9ons
ICWSM
2012
Y.
Volkovich†,
S.
Scellato‡,
D.
Laniado†,
C.
Mascolo‡,
A.
Kaltenbrunner†
†
Barcelona
Media
Founda9on
‡
University
of
Cambridge
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
1
2. Background
• Geographically
closer
pair
of
individuals
are
more
likely
to
develop
social
bonds
– [Merton,
1948],
[Fes9nger
et
al.,
1950],
…
• This
holds
even
on
online
social
networking
services
– [Liben-‐Nowell
et
al.,
2005],
[Backstorm
et
al.,
2010],
…
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
2
3. Research
Ques9on
What
is
the
rela;onship
between
the
structural
proper,es
of
online
social
;es
and
the
spa,al
distance
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
3
4. Structure,
Tie
strength,
and
Distance
core
• Tie
strength
– Close
friends
or
just
acquaintances
– Frequency
of
interac9ons
bridge
• Network
structure
– Cores,
Bridges,
Periphery
strong
9e
• Spa9al
distance
– Between
connected
individuals
periphery
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
4
5. Results
• Individuals
at
closer
distance
more
likely
to
establish
social
connec9ons
• Social
links
in
the
core
tend
to
span
shorter
distances
than
outer
9es
• Interac9on
levels
are
higher
inside
the
core
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
5
6. Dataset
“Spanish
Facebook”
Founded
in
2006
in
Spain
Full
anonymized
snapshot
as
of
Nov.
2010
p 9.8
million
users
(25%
of
Spanish
popula9on)
p 580
million
friendship
links
p 500
million
message
exchanges
in
3
months
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
6
7. Basic
Proper9es
N:
#
of
nodes
deff:
90%
diameter
K:
#
of
edges
dmax:
Maximal
distance
NGC:
#
of
nodes
in
<dpath>:
Average
path-‐length
giant
component
<D>:
Average
spa9al
distance
<deg>:
Average
degree
of
arbitrary
pairs
<C>:
Clustering
coefficient
<l>:
Average
spa9al
distance
of
connected
pairs
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
7
8. Structural
posi9on
of
social
9es
(1/2)
• Local
posi9on
(
social
overlap
)
– Connected
users
with
many
common
friends
seem
to
be
inside
the
core
oi, j = Γ i ∩ Γ j
Γi :
set
of
friends
of
i
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
8
9. Structural
posi9on
of
social
9es
(2/2)
• Global
posi9on
(
k-‐index
)
– K-‐index
measures
how
middle
a
node
is
in
the
network
– k-‐index
of
a
node
is
v
if
it
belongs
to
v-‐core
but
not
The
k-‐core
is
the
maximal
subgraph
to
(v+1)-‐core
where
each
node
connects
to
at
least
k
nodes
inside
the
subgraph.
– K-‐index
of
an
edge
eij
kij = min(ki , k j )
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
9
10. Structural
posi9on
and
spa9al
length
(1/2)
Spa,al
distance
decreases
as
connected
users
share
more
friends
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
10
11. Structural
posi9on
and
spa9al
length
(2/2)
Social
links
inside
the
core
are
shorter
than
the
ones
reaching
the
periphery
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
11
12. The
impact
of
9e
strength
(1/2)
Although
likelihood
of
friendship
is
correlated
to
the
spa,al
distance,
,e
strength
does
not
affect
the
distance
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
12
13. The
impact
of
9e
strength
(2/2)
p Users
inside
the
core
more
frequently
interact
p Users
in
the
periphery
and
in
the
core
more
frequently
interact
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
13
14. Discussion
• Shorter
distance
increase
the
likelihood
that
users
belong
to
the
same
dense
group.
• Bridges
create
not
only
network
shortcuts,
but
also
spa9al
shortcuts.
• Frequent
interac9ons
occur
inside
the
cores
or
between
the
core
and
periphery.
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
14
15. Conclusion
• Inves9gated
how
spa9al
constraints
influence
the
network
structure
Spa9al
Tie
core
distance
strength
Spa9al
distance
high
Tie
strength
low
core
low
high
13/02/18
KDE
Seminar:
Yuto
Yamaguchi
15