The Role of Social Networks in Information
Diusion
Eytan Bakshy1;2;y, Itamar Rosenn1, Cameron Marlow1, Lada Adamic2
1 Facebook, Menlo Park, CA, USA
2 University of Michigan, Ann Arbor, MI, USA
y Corresponding author: ebakshy@fb.com
Abstract
Online social networking technologies enable individuals to simultane-
ously share information with any number of peers. Quantifying the causal
eect of these technologies on the dissemination of information requires
not only identication of who in
uences whom, but also of whether indi-
viduals would still propagate information in the absence of social signals
about that information. We examine the role of social networks in online
information diusion with a large-scale eld experiment that randomizes
exposure to signals about friends' information sharing among 253 million
subjects in situ. Those who are exposed are signicantly more likely to
spread information, and do so sooner than those who are not exposed.
We further examine the relative role of strong and weak ties in informa-
tion propagation. We show that, although stronger ties are individually
more in
uential, it is the more abundant weak ties who are responsible
for the propagation of novel information. This suggests that weak ties
may play a more dominant role in the dissemination of information online
than currently believed.
1 Introduction
Social in
uence can play a crucial role in a range of behavioral phenomena,
from the dissemination of information, to the adoption of political opinions and
health-related behaviors [22, 40, 16], which are increasingly mediated through
online systems [36, 17]. Despite the wide availability of data from online social
networks, identifying in
uence remains a challenge. Individuals tend to engage
in similar activities as their peers, so it is often impossible to determine from
observational data whether a correlation between two individuals' behaviors ex-
ists because they are similar or because one person's behavior has in
uenced
the other [5, 31, 37]. In the context of information diusion, two people may
disseminate the same information as each other because they possess the same
information sources, such as web sites or television, that they consume regu-
larly [3, 36].
Moreover, homophily { the tendency of individuals with similar characteris-
tics to associate with one another [33, 1, 27] { creates diculties for measuring
the relative role of strong and weak ties in information diusion, since peo-
ple are more similar to those with whom they interact often [21, 33]. On one
hand, pairs of individuals who interact more often have greater opportunity to
in
uence one another and have more aligned interests, increasing the chances
of contagion [11, 26]. However, this commonality amplies the potential for
confounds: those who interact more often are more likely to have increasingly
1
ar
X
iv
:1
20
1.
41
45
v1
[
cs
.SI
]
19
Ja
n 2
01
2

similar information sources. As a result, inferences made from observational
data may overstate the importance of strong ties in information spread. Con-
versely, individuals who interact infrequently have more diverse social networks
that provide access to novel information [21, 12]. But because contact between
such ties is intermittent, and the individuals tend to be dissimilar, any particu-
lar piece of information is less likely to
ow across weak ties [14, 35]. Historical
attempts to collect data on how often pairs of individuals communicate and
where they get their information have been prone to biases [32, 10], further
obscuring the empirical relationship between tie strength and diusion.
Confounding factors related to homophily can be addressed using controlled
experiments, but experimental work has thus far been conned to the spread
of highly specic information within limited populations [13, 6]. In order to
understand how information spreads in a real-world environment, we wish to
examine a setting where a large population of individuals frequently exchange
information with their peers. Facebook is the most widely used social network-
ing service in the world, with over 800 million people using the service each
month. For example, in the United States, 54% of adult Internet users are
on Facebook [25]. Those American users on average maintain 48% of their real
world contacts on the site [25], and many of these individuals regularly exchange
news items with their contacts [36]. In addition, interaction among users is well
correlated with self-reported intimacy [18]. Thus, Facebook represents a broad
online population of individuals whose online personal networks re
ect their
real-world connections, making it an ideal environment to study information
contagion.
We use an experimental approach on Facebook to measure the spread of in-
formation sharing behaviors. The experiment randomizes whether individuals
are exposed via Facebook to information about their friends' sharing behavior,
thereby devising two worlds under which information spreads: one in which
certain information can only be acquired external to Facebook, and another in
which information can be acquired within or external to Facebook. By compar-
ing the behavior of individuals within these two conditions, we can determine
the causal eect of the medium on information sharing.
The remainder of this paper is organized as follows. We further motivate
our study with additional related work in Section 2. Our experimental design is
described in Section 3. Then, in Section 4 we discuss the causal eect of exposure
to content on the newsfeed, and how friends' sharing behavior is correlated in
time, irrespective of social in
uence via the newsfeed. Furthermore, we show
that multiple sharing friends are predictive of sharing behavior regardless of
exposure on the feed, and that additional friends do indeed have an increasing
causal eect on the propensity to share. In Section 5 we discuss how tie strength
relates to in
uence and information diusion. We show that users are more
likely to have the same information sources as their close friends, and that
simultaneously, these close friends are more likely to in
uence subjects. Using
the empirical distribution of tie strength in the network, we go on to compute
the overall eect of strong and weak ties on the spread of information in the
network. Finally, we discuss the implications of our work in Section 6.
2