Information Diffusion in Online Social Networks:

A Survey
Adrien Guille
1
Hakim Hacid
2
Ccile Favre
1
Djamel A. Zighed
1,3
1
ERIC Lab, Lyon 2 University, France
{rstname.lastname}@univ-lyon2.fr
2
Bell Labs France, Alcatel-Lucent, France
[email protected]
3
Institute of Human Science, Lyon 2 University, France
[email protected]
ABSTRACT
Online social networks play a major role in the spread of
information at very large scale. A lot of effort have been
made in order to understand this phenomenon, rang-
ing from popular topic detection to information diffu-
sion modeling, including inuential spreaders identi-
cation. In this article, we present a survey of represen-
tative methods dealing with these issues and propose a
taxonomy that summarizes the state-of-the-art. The ob-
jective is to provide a comprehensive analysis and guide
of existing efforts around information diffusion in social
networks. This survey is intended to help researchers in
quickly understanding existing works and possible im-
provements to bring.
1. INTRODUCTION
Online social networks allow hundreds of millions
of Internet users worldwide to produce and con-
sume content. They provide access to a very vast
source of information on an unprecedented scale.
Online social networks play a major role in the dif-
fusion of information by increasing the spread of
novel information and diverse viewpoints [3]. They
have proved to be very powerful in many situations,
like Facebook during the 2010 Arab spring [22] or
Twitter during the 2008 U.S. presidential elections
[23] for instance. Given the impact of online social
networks on society, the recent focus is on extract-
ing valuable information from this huge amount of
data. Events, issues, interests, etc. happen and
evolve very quickly in social networks and their cap-
ture, understanding, visualization, and prediction
are becoming critical expectations from both end-
users and researchers. This is motivated by the fact
that understanding the dynamics of these networks
may help in better following events (e.g. analyz-
ing revolutionary waves), solving issues (e.g. pre-
venting terrorist attacks, anticipating natural haz-
ards), optimizing business performance (e.g. opti-
mizing social marketing campaigns), etc. Therefore
researchers have in recent years developed a vari-
ety of techniques and models to capture informa-
tion diusion in online social networks, analyze it,
extract knowledge from it and predict it.
Information diusion is a vast research domain
and has attracted research interests from many elds,
such as physics, biology, etc. The diusion of in-
novation over a network is one of the original rea-
sons for studying networks and the spread of disease
among a population has been studied for centuries.
As computer scientists, we focus here on the par-
ticular case of information diusion in online so-
cial networks, that raises the following questions :
(i) which pieces of information or topics are popu-
lar and diuse the most, (ii) how, why and through
which paths information is diusing, and will be dif-
fused in the future, (iii) which members of the net-
work play important roles in the spreading process?
The main goal of this paper is to review develop-
ments regarding these issues in order to provide a
simplied view of the eld. With this in mind, we
point out strengths and weaknesses of existing ap-
proaches and structure them in a taxonomy. This
study is designed to serve as guidelines for scien-
tists and practitioners who intend to design new
methods in this area. This also will be helpful for
developers who intend to apply existing techniques
on specic problems since we present a library of
existing approaches in this area.
The rest of this paper is organized as follows.
In Section 2 we detail online social networks basic
characteristics and information diusion properties.
In Section 3 we present methods to detect topics of
interest in social networks using information diu-
sion properties. Then we discuss how to model in-
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
Author manuscript, published in "Sigmod Record 42, 2 (2013) 17-28"
DOI : 10.1145/2503792.2503797
formation diusion and detail both explanatory and
predictive models in Section 4. Next, we present
methods to identify inuential information spread-
ers in Section 5. In the last section we summarize
the reviewed methods in a taxonomy, discuss their
shortcomings and indicate open questions.
2. BASICS OF ONLINE SOCIAL NET-
WORKS ANDINFORMATIONDIFFU-
SION
An online social network (OSN) results from the
use of a dedicated web-service, often referred to as
social network site (SNS), that allows its users to (i)
create a prole page and publish messages, and (ii)
explicitly connect to other users thus creating social
relationships. De facto, an OSN can be described
as a user-generated content system that permits its
users to communicate and share information.
An OSN is formally represented by a graph, where
nodes are users and edges are relationships that can
be either directed or not depending on how the SNS
manages relationships. More precisely, it depends
on whether it allows connecting in an unilateral
(e.g. Twitter social model of following) or bilateral
(e.g. Facebook social model of friendship) manner.
Messages are the main information vehicle in such
services. Users publish messages to share or for-
ward various kinds of information, such as product
recommendations, political opinions, ideas, etc. A
message is described by (i) a text, (ii) an author,
(iii) a time-stamp and optionally, (iv) the set of
people (called mentioned users in the social net-
working jargon) to whom the message is specically
targeted. Figure 1 shows an OSN represented by a
directed graph enriched by the messages published
by its four members. An arc e = (u
x
, u
y
) means
that the user u
x
is exposed to the messages pub-
lished by u
y
. This representation reveals that,
for example, the user named u
1
is exposed to the
content shared by u
2
and u
3
. It also indicates
that no one receives the messages written by u
4
.
DEFINITION 1 (Topic). A coherent set of
semantically related terms that express a single ar-
gument. In practice, we nd three interpretations
of this denition: (i) a set S of terms, with |S| = 1,
e.g. {obama} (ii) a set S of terms, with |S| > 1,
e.g. {obama, visit, china} and (iii) a proba-
bility distribution over a set S of terms.
Every piece of information can be transformed
into a topic [6, 30] using one of the common for-
malisms detailed in Denition 1. Globally, the con-
tent produced by the members of an OSN is a stream
of messages. Figure 2 represents the stream pro-
duced by the members of the network depicted in
the previous example. That stream can be viewed
as a sequence of decisions (i.e. whether to adopt
a certain topic or not), with later people watching
the actions of earlier people. Therefore, individuals
are inuenced by the actions taken by others. This
eect is known as social inuence [2], and is dened
as follows:
DEFINITION 2 (Social Influence). A so-
cial phenomenon that individuals can undergo or ex-
ert, also called imitation, translating the fact that
actions of a user can induce his connections to be-
have in a similar way. Inuence appears explicitly
when someone retweets someone else for example.
DEFINITION 3 (Herd behavior). A social
behavior occurring when a sequence of individuals
make an identical action, not necessarily ignoring
their private information signals.
DEFINITION 4 (Information Cascade).
A behavior of information adoption by people in a
social network resulting from the fact that people
ignore their own information signals and make de-
cisions from inferences based on earlier peoples ac-
tions.
u
1
u
3
u
2
m
1
m
2
m
3
m
4
m
5
m
6
u
4
m
7
Figure 1: An example of OSN enriched by
users messages. Users are denoted u
i
and
messages m
j
. An arc (u
x
, u
y
) means that u
x
is exposed to the messages published by u
y
.
m
1
m
3
m
2
m
7
m
6
m
5
m
4
time
Figure 2: The stream of messages produced
by the members of the network depicted on
Figure 1.
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
Based on the social inuence eect, information
can spread across the network through the prin-
ciples of herd behavior and informational cascade
which we dene respectively in Denition 3 and 4.
In this context, some topics can become extremely
popular, spread worldwide, and contribute to new
trends. Eventually, the ingredients of an informa-
tion diusion process taking place in an OSN can
be summarized as follows: (i) a piece of information
carried by messages, (ii) spreads along the edges
of the network according to particular mechanics,
(iii) depending on specic properties of the edges
and nodes. In the following sections, we will dis-
cuss these dierent aspects with the most relevant
recent work related to them as well as an analysis
of weaknesses, strength, and possible improvements
for each aspect.
3. DETECTING POPULAR TOPICS
One of the main tasks when studying information
diusion is to develop automatic means to provide
a global view of the topics that are popular over
time or will become popular, and animate the net-
work. This involves extracting tables of content
to sum up discussions, recommending popular top-
ics to users, or predicting future popular topics.
Traditional topic detection techniques developed
to analyze static corpora are not adapted to mes-
sage streams generated by OSNs. In order to e-
ciently detect topics in textual streams, it has been
suggested to focus on bursts. In his seminal work,
Kleinberg [26] proposes a state machine to model
the arrival times of documents in a stream in or-
der to identify bursts, assuming that all the docu-
ments belong to the same topic. Leskovec et al. [27]
show that the temporal dynamics of the most pop-
ular topics in social media are indeed made up of a
succession of rising and falling patterns of popular-
ity, in other words, successive bursts of popularity.
Figure 3 shows a typical example of the temporal
dynamics of top topics in OSNs.
DEFINITION 5 (Bursty topic). A behav-
ior associated to a topic within a time interval in
which it has been extensively treated but rarely be-
fore and after.
In the following, we detail methods designed to
detect topics that have drawn bursts of interest, i.e.
bursty topics (see Denition 5), from a stream of
topically diverse messages.
All approaches detailed hereafter rely on the com-
putation of some frequencies and work on discrete
data. Therefore they require the stream of mes-
sages to be discretized. This is done by transform-
time
I
e
v
e
I

o
f

a
t
t
e
n
t
i
o
n
Figure 3: Temporal dynamics of popular top-
ics. Each shade of gray represents a topic.
ing the raw continuous data into a sequence of col-
lection of messages published during equally sized
time slices. This principle is illustrated on Figure 4,
which shows a possible discretization of the stream
previously depicted in Figure 2. This pre-processing
step is not trivial since it denes the granularity of
the topic detection. A very ne discretization (i.e.
short time-slices) will allow to detect topics that
were popular during short periods whereas a dis-
cretization using longer time-slices will not.
m
1
m
3 m
2
m
7
m
4
m
5
m
6
time-sIice 1 2
Figure 4: A possible discretization of the
stream of messages shown on Figure 2.
Shamma et al. [46] propose a simple model, PT
(i.e. Peaky Topics) , similar to the classical tf-idf
model [44] in the sense that it is based on a normal-
ized term frequency metric. In order to quantify the
overall term usage, they consider each time slice as
a pseudo-document composed of all the messages in
the corresponding collection. The normalized term
frequency ntf is dened as follows: ntf
t,i
=
tft,i
cft
,
where tf
t,i
is the frequency of term t at the i
th
time
slice and cf
t
is the frequency of term t in the whole
message stream. Using that metric, bursty topics
dened as single terms are ranked. However, some
terms can be polysemous or ambiguous and a single
term doesnt seem to be enough to clearly identify a
topic. Therefore, more sophisticated methods have
been developed.
AlSumait et al. [1] propose an online topic model,
more precisely, a non-Markov on-line LDA Gibbs
sampler topic model, called OLDA. Basically, LDA
(i.e. Latent Dirichlet Allocation [4]) is a statis-
tical generative model that relies on a hierarchi-
cal Bayesian network that relates words and mes-
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
sages through latent topics. The generative process
behind is that documents are represented as ran-
dom mixtures over latent topics, where each topic
is characterized by a distribution over words. The
idea of OLDA is to incrementally update the topic
model at each time slice using the previously gen-
erated model as a prior and the corresponding col-
lection of messages to guide the learning of the new
generative process. This method builds an evolu-
tionary matrix for each topic that captures the evo-
lution of the topic over time and thus permits to
detect bursty topics.
Cataldi et al. [6] propose the TSTE method (i.e.
Temporal and Social Terms Evaluation) that con-
siders both temporal and social properties of the
stream of messages. To this end, they develop a
ve-step process that rstly formalize the messages
content as vectors of terms with their relative fre-
quencies computed by using the augmented normal-
ized term frequency [43]. Then, the authority of
the active authors is assessed using their relation-
ships and the Page Rank algorithm [35]. It allows
to model the life cycle of each term on the basis of a
biological metaphor, which is based on the calcula-
tion of values of nutrition and energy that leverage
the users authority. Using supervised or unsuper-
vised techniques, rooted in the calculation of a crit-
ical drop value based on the energy, the proposed
method can identify most bursty terms. Finally, a
solution is provided to dene bursty topics as sets
of terms using a co-occurence based metric.
These methods identify particular topics that have
drawn bursts of interest in the past. Lu et al. [40]
develop a method that permits predicting which
topics will draw attention in the near future. Au-
thors propose to adapt a technical analysis indi-
cator primary used for stock price study, namely
MACD (i.e. Moving Average Convergence Diver-
gence), to identify bursty topics, dened as a single
term. The principle of MACD is to turn two trend-
following indicators, precisely a short period and a
longer period moving average of terms frequency,
into a momentum oscillator. The trend momentum
is obtained by calculating the dierence between the
long and the shorter moving averages. Authors give
two simple rules to identify when the trends of a
term will rise: (i) when the value of the trend mo-
mentum changes from negative to positive, the topic
is beginning to rise; (ii) when the value changes from
positive to negative, the level of attention given to
the topic is falling.
The above methods are based on the detection
of unusual term frequencies in exchanged messages
to detect interesting topics in OSNs. However, more
and more frequently, OSNs users publish non-textual
content such as URL, pictures or videos. To deal
with non-textual content, Takahashi et al. [47] pro-
pose to use mentions contained in messages to iden-
tify bursty topics, instead of focusing on the textual
content. Mentioning is a social practice used to ex-
plicitly target messages and eventually engage dis-
cussion. For that, they develop a method that com-
bines a mentioning anomaly score and a change-
point detection technique based on SDNML (i.e.
Sequentially Discounting Normalized Maximum Like-
lihood). The anomaly is calculated with respect
to the standard mentioning behavior of each user,
which is estimated by a probability model.
Table 1 summarizes the surveyed methods ac-
cording to four axes. The table is structured ac-
cording to four main criteria that allow for a quick
comparison: (i) how is a topic dened, (ii) which
dimensions are incorporated into each method, (iii)
which types of content each method can handle, and
(iv) either the method detects actual bursts or pre-
dicts them. It should be noted that the table is not
intended to express any preference regarding one
method or another, but rather to present a global
comparison.
r
e
f
e
r
e
n
c
e
t
o
p
i
c
d
e

n
i
t
i
o
n
d
i
m
e
n
s
i
o
n
(
s
)
c
o
n
t
e
n
t
t
y
p
e
t
a
s
k
t
y
p
e
s
i
n
g
l
e
t
e
r
m
s
e
t
o
f
t
e
r
m
s
d
i
s
t
r
i
b
u
t
i
o
n
c
o
n
t
e
n
t
s
o
c
i
a
l
t
e
x
t
u
a
l
n
o
n
-
t
e
x
t
u
a
l
o
b
s
e
r
v
a
t
i
o
n
p
r
e
d
i
c
t
i
o
n
PT x x x x
OLDA x x x x
TSTE x x x x x
SDNML x x x x x
MACD x x x x
Table 1: Summary of topic detection ap-
proaches w.r.t topic denition, incorporated
dimensions, handled content and the task.
4. MODELINGINFORMATIONDIFFU-
SION
Modeling how information spreads is of outstand-
ing interest for stopping the spread of viruses, ana-
lyzing how misinformation spread, etc. In this sec-
tion, we rst give the basics of diusion modeling
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
and then detail the dierent models proposed to
capture or predict spreading processes in OSNs.
DEFINITION 6 (Activation Sequence). .
An ordered set of nodes capturing the order in which
the nodes of the network adopted a piece of infor-
mation.
DEFINITION 7 (Spreading Cascade). A
directed tree having as a root the rst node of the
activation sequence. The tree captures the inuence
between nodes (branches represent who transmitted
the information to whom) and unfolds in the same
order as the activation sequence.
The diusion process is characterized by two as-
pects: its structure, i.e. the diusion graph that
transcribes who inuenced whom, and its temporal
dynamics, i.e. the evolution of the diusion rate
which is dened as the amount of nodes that adopts
the piece of information over time. The simplest
way to describe the spreading process is to consider
that a node can be either activated (i.e. has re-
ceived the information and tries to propagate it) or
not. Thus, the propagation process can be viewed
as a successive activation of nodes throughout the
network, called activation sequence, dened in Def-
inition 6.
Usually, models developed in the context of OSNs
assume that people are only inuenced by actions
taken by their connections. To put it dierently,
they consider that an OSN is a closed world and
assume that information spreads because of infor-
mational cascades. That is why the path followed
by a piece of information in the network (i.e. the
diusion graph) is often referred to as the spread-
ing cascade, dened in Denition 7. Activation se-
quences are simply extracted from data by collect-
ing messages dealing with the studied information,
i.e. topic, and ordering them according to the time
axis. This principle is illustrated in Figure 5. It
provides knowledge about where and when a piece
of information propagated but not how and why did
it propagate. Therefore, there is a need for models
that can capture and predict the hidden mechanism
underlying diusion. We can distinguish two cate-
gories of models in this scope: (i) explanatory mod-
els and (ii) predictive models. In the following, we
detail these two categories and analyze some repre-
sentative eorts in both of them.
4.1 Explanatory Models
The aim of explanatory models is to infer the un-
derlying spreading cascade, given a complete acti-
vation sequence. These models make it possible to
retrace the path taken by a piece of information
u
1
u
3
u
2
m
1
m
2
m
3
m
4
m
5
m
6
u
4
m
7
m
1
m
3
m
2
m
7
m
6
m
5
m
4
time
t
2
t
1
t
3
t
4
u
2
u
3
u
5
u
4
u
1
u
6
u
7
Figure 5: An OSN in which darker nodes
took part in the diusion process of a par-
ticular information. The activation sequence
can be extracted using the time at which the
messages were published: [u
4
; u
2
; u
3
; u
5
], with
t
1
< t
2
< t
3
< t
4
.
and are very useful to understand how information
propagated.
Gomez et al. [15] propose to explore correla-
tions in nodes infections times to infer the struc-
ture of the spreading cascade and assume that acti-
vated nodes inuence each of their neighbors inde-
pendently with some probability. Thus, the proba-
bility that one node had transmitted information to
another is decreasing in the dierence of their ac-
tivation time. They develop NETINF, an iterative
algorithm based on submodular function optimiza-
tion for nding the spreading cascade that maxi-
mizes the likelihood of observed data.
Gomez et al. [14] extend NETINF and propose
to model the diusion process as a spatially dis-
crete network of continuous, conditionally indepen-
dent temporal processes occurring at dierent rates.
The likelihood of a node infecting another at a given
time is modeled via a probability density function
depending on infection times and the transmission
rate between the two nodes. The proposed algo-
rithm, NETRATE, infers pairwise transmission rates
and the graph of diusion by formulating and solv-
ing a convex maximum likelihood problem [9].
These methods consider that the underlying net-
work remains static over time. This is not a satisfy-
ing assumption, since the topology of OSNs evolves
very quickly, both in terms of edges creation and
deletion. For that reason, Gomez et al. [16] extend
NETRATE and propose a time-varying inference
algorithm, INFOPATH, that uses stochastic gradi-
ents to provide on-line estimates of the structure
and temporal dynamics of a network that changes
over time.
In addition, because of technical and crawling
API limitations, there is a data acquisition bottle-
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
r
e
f
e
r
e
n
c
e
n
e
t
w
o
r
k
i
n
f
e
r
r
e
d
p
r
o
p
e
r
t
i
e
s
s
u
p
p
o
r
t
s
m
i
s
s
i
n
g
d
a
t
a
s
t
a
t
i
c
d
y
n
a
m
i
c
p
a
i
r
w
a
i
s
e
t
r
a
n
s
m
i
s
s
i
o
n
p
r
o
b
a
b
i
l
i
t
y
p
a
i
r
w
a
i
s
e
t
r
a
n
s
m
i
s
s
i
o
n
r
a
t
e
c
a
s
c
a
d
e
p
r
o
p
e
r
t
i
e
s
NETINF x x x
NETRATE x x x x
INFOPATH x x x x x
k-tree model x x x
Table 2: Summary of explanatory models
w.r.t the nature of the underlying network,
inferred properties and the ability of the
method to work with incomplete data.
neck potentially responsible for missing data. To
overcome this issue, one approach is to crawl data
as eciently as possible. Choudhury et al. [7]
analysed how the data sampling strategy impacts
the discovery of information diusion in social me-
dia. Based on experimentations on Twitter data,
they concluded that sampling methods that con-
sider both network topology and users attributes
such as activity and localisation allow to capture
information diusion with lower error in compari-
son to naive strategies, like random or activity-only
based sampling. Another approach is to develop
specic models that assume that data are missing.
Sadikov et al. [41] develop a method based on a k-
tree model designed to, given only a fraction of the
complete activation sequence, estimate the proper-
ties of the complete spreading cascade, such as its
size or depth.
We summarize the surveyed explanatory models
in Table 2. In the following, we detail the second
category of models, namely, predictive models.
4.2 Predictive Models
These models aim at predicting how a specic dif-
fusion process would unfold in a given network, from
temporal and/or spatial points of view by learning
from past diusion traces. We classify existing mod-
els into two development axes, graph and non-graph
based approaches.
step 0: initiaI
users
step 1
step 2
step 3
step 4
Figure 6: A spreading process modeled by
Independent Cascades in four steps.
4.2.1 Graph based approaches
There are two seminal models in this category,
namely Independent Cascades (IC) [13] and Linear
Threshold (LT) [17]. They assume the existence
of a static graph structure underlying the diusion
and focus on the structure of the process. They
are based on a directed graph where each node can
be activated or not with a monotonicity assump-
tion, i.e. activated nodes cannot deactivate. The IC
model requires a diusion probability to be associ-
ated to each edge whereas LT requires an inuence
degree to be dened on each edge and an inuence
threshold for each node. For both models, the dif-
fusion process proceeds iteratively in a synchronous
way along a discrete time-axis, starting from a set
of initially activated nodes, commonly named early
adopters [37]:
DEFINITION 8 (Early Adopters). A set
of users who are the rst to adopt a piece of in-
formation and then trigger its diusion.
In the case of IC, for each iteration, the newly
activated nodes try once to activate their neigh-
bors with the probability dened on the edge joining
them. In the case of LT, at each iteration, the in-
active nodes are activated by their activated neigh-
bors if the sum of inuence degrees exceeds their
own inuence threshold. Successful activations are
eective at the next iteration. In both cases, the
process ends when no new transmission is possible,
i.e. no neighboring node can be contacted. These
two mechanisms reect two dierent points of view:
IC is sender-centric while LT is receiver-centric. An
example of spreading process modeled with IC is
given by Figure 6. We detail hereafter models aris-
ing from those approaches and adapted to OSNs.
Galuba et al. [11] propose to use the LT model
to predict the graph of diusion, having already ob-
served the beginning of the process. Their model
relies on parameters such as information virality,
pairwise users degree of inuence and user proba-
bility of adopting any information. The LT model
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
is tted on the data describing the beginning of the
diusion process by optimizing the parameters us-
ing the gradient ascent method. However, LT cant
reproduce realistic temporal dynamics.
Saito et al. [42] relax the synchronicity assump-
tion of traditional IC and LT graph-based mod-
els by proposing asynchronous extensions. Named
AsIC and AsLT (i.e. asynchronous independent
cascades and asynchronous linear threshold), they
proceed iteratively along a continuous time axis and
require the same parameters as their synchronous
counterparts plus a time-delay parameter on each
edge of the graph. Models parameters are dened
in a parametric way and authors provide a method
to learn the functional dependency of the model
parameters from nodes attributes. They formulate
the task as a maximum likelihood estimation prob-
lem and an update algorithm that guarantees the
convergence is derived. However, they only exper-
imented with synthetic data and dont provide a
practical solution.
Guille et al. [19] also model the propagation pro-
cess as asynchronous independent cascades. They
develop theT-BaSIC model (i.e. Time-Based Asyn-
chronous Independent Cascades), which parameters
arent xed numerical values but functions depend-
ing on time. The model parameters are estimated
from social, semantic and temporal nodes features
using logistic regression.
4.2.2 Non-graph based approaches
Non-graph based approaches do not assume the
existence of a specic graph structure and have been
mainly developed to model epidemiological processes.
They classify nodes into several classes (i.e. states)
and focus on the evolution of the proportions of
nodes in each class. SIR and SIS are the two sem-
inal models [21, 34], where S stands for suscepti-
ble, I for infected (i.e. adopted the information)
and R for recovered (i.e. refractory). In both cases,
nodes in the S class switch to the I class with a xed
probability . Then, in the case of SIS, nodes in
the I class switch to the S class with a xed prob-
ability , whereas in the case of SIR they perma-
nently switch to the R class. The percentage of
nodes in each class is expressed by simple dier-
ential equations. Both models assume that every
node has the same probability to be connected to
another and thus connections inside the population
are made at random.
Leskovec et al. [28] propose a simple and intu-
itive SIS model that requires a single parameter,
. It assumes that all nodes have the same prob-
ability to adopt the information and nodes that

u u u
1 2 3
u
1
u
2
u
3
I
I
I
v
o
I
u
m
e
time t t t

t
t-t

t-t

t-t

Figure 7: LIM forecasts the rate of diu-

sion by summing the inuence functions of
a given set of early adopters. Here, the early
adopters are u
1
, u
2
and u
3
whose respective
inuence functions are Iu
1
, Iu
2
and Iu
3
.
have adopted the information become susceptible
at the next time-step (i.e. = 1). This is a strong
assumption since in real-world social networks, in-
uence is not evenly distributed between all nodes
and it is necessary to develop more complex mod-
eling that take into account this characteristic.
Yang et al. [50] start from the assumption that
the diusion of information is governed by the in-
uence of individual nodes. The method focuses
on predicting the temporal dynamics of information
diusion, under the form of a time-series describ-
ing the rate of diusion of a piece of information,
i.e. the volume of nodes that adopt the informa-
tion through time. They develop a Linear Inu-
ence model (LIM), where the inuence functions
of individual nodes govern the overall rate of dif-
fusion. The inuence functions are represented in
a non-parametric way and are estimated by solv-
ing a non-negative least squares problem using the
Reective Newton Method [8]. Figure 7 illustrates
how LIM forecasts the rate of diusion from a set
of early adopters and their activation time.
Wang et al. [48] propose a Partial Dierential
Equation (PDE) based model to predict the diu-
sion of an information injected in the network by a
given node. More precisely, a diusive logistic equa-
tion model is used to predict both topological and
temporal dynamics. Here, the topology of the net-
work is considered only in term of the distance from
each node to the source node. The dynamics of the
process is given by a logistic equation that models
the density of inuenced users at a given distance of
the source and at a given time. That denition of
the network topology allows to formulate the prob-
lem simply, as for classical non-graph based meth-
ods while integrating some spatial knowledge. The
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
r
e
f
e
r
e
n
c
e
d
i
m
e
n
s
i
o
n
(
s
)
b
a
s
i
s
m
a
t
h
e
m
a
t
i
c
a
l
m
o
d
e
l
i
n
g
s
o
c
i
a
l
t
i
m
e
c
o
n
t
e
n
t
g
r
a
p
h
b
a
s
e
d
n
o
n
-
g
r
a
p
h
b
a
s
e
d
p
a
r
a
m
e
t
r
i
c
n
o
n
-
p
a
r
a
m
e
t
r
i
c
LT-based x x x x
AsIC, n/a n/a n/a x x
AsLT
T-BaSIC x x x x x
SIS-based x x x
LIM x x x x
PDE x x x x
Table 3: Summary of diusion predic-
tion methods, distinguishing graph and non-
graph based approaches w.r.t incorporated
dimensions and mathematical modeling.
parameters of the model are estimated using the
Cubic Spline Interpolation method [12].
We summarize the surveyed predictive models in
Table 3. In the following section, we discuss the
role of nodes in the propagation process and how to
identify inuential spreaders.
5. IDENTIFYINGINFLUENTIALINFOR-
MATION SPREADERS
Identifying the most inuential spreaders in a net-
work is critical for ensuring ecient diusion of in-
formation. For instance, a social media campaign
can be optimized by targeting inuential individuals
who can trigger large cascades of further adoptions.
This section presents briey some methods that il-
lustrate the various possible ways to measure the
relative importance and inuence of each node in
an online social network.
DEFINITION 9 (K-Core). Let G be a graph.
If H is a sub-graph of G, (H) will denote the min-
imum degree of H. Thus each node of H is adja-
cent to at least (H) other nodes of H. If H is a
maximal connected (induced) sub-graph of G with
(H) >= k, we say that H is a k-core of G [45].
Kitsak et al. [25] show that the best spreaders
are not necessarily the most connected people in the
network. They nd that the most ecient spreaders
are those located within the core of the network as
identied by the k-core decomposition analysis [45],
as dened in Denition 9. Basically, the principle of
the k-core decomposition is to assign a core index k
s
to each node such that nodes with the lowest values
are located at the periphery of the network while
nodes with the highest values are located in the
center of the network. The innermost nodes thus
forms the core of the network. Brown et al. [5] ob-
serve that the results of the k-shell decomposition
on Twitter network are highly skewed. Therefore
they propose a modied algorithm that uses a log-
arithmic mapping, in order to produce fewer and
more meaningful k-shell values.
Cataldi et al. [6] propose to use the well known
PageRank algorithm [35] to assess the distribution
of inuence throughout the network. The PageR-
ank value of a given node is proportional to the
probability of visiting that node in a random walk
of the social network, where the set of states of the
random walk is the set of nodes.
The methods we have just described only exploit
the topology of the network, and ignore other im-
portant properties, such as nodes features and the
way they process information. Starting from the
observation that most OSNs members are passive
information consumers, Romero et al. [38] develop
a graph-based approach similar to the well known
HITS algorithm, IP (i.e. Inuence-Passivity), that
assigns a relative inuence and a passivity score
to every users based on the ratio at which they
forward information. However, no individual can
be a universal inuencer, and inuential members
of the network tend to be inuential only in one
or some specic domains of knowledge. Therefore,
Pal et al. [36] develop a non-graph based, topic-
sensitive method. To do so, they dene a set of
nodal and topical features for characterizing the
network members. Using probabilistic clustering
over this feature space, they rank nodes with a
within-cluster ranking procedure to identify the most
inuential and authoritative people for a given topic.
Weng et al. [49] also develop a topic-sensitive ver-
sion of the Page Rank algorithm dedicated to Twit-
ter, TwitterRank.
Kempe et al. [24] adopt a dierent approach and
propose to use the IC and LT models (previously
described in Section 4.2.1) to tackle the inuence
maximization problem. This problem asks, for a
parameter k, to nd a k-node set of maximum in-
uence in the network. The inuence of a given
set of nodes corresponds to the number of activated
nodes at the end of the diusion process according
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
r
e
f
e
r
e
n
c
e
g
r
a
p
h
b
a
s
e
d
i
n
c
o
r
p
o
r
a
t
e
d
d
i
m
e
n
s
i
o
n
(
s
)
u
s
e
r
s

f
e
a
t
u
r
e
s
t
o
p
i
c
k-shell decomposition x
log k-shell decomposition x
PageRank x
Topic-sensitive PageRank x x
IP x x
Topical Authorities x x
k-node set x
Table 4: Summary of inuential spreaders
identication methods distinguishing graph
and non-graph based approaches w.r.t incor-
porated dimensions.
to IC or LT, using this set as the set of initially
activated nodes. They provide an approximation
for this optimization problem using a greedy hill-
climbing strategy based on submodular functions.
The surveyed inuence assessment methods are
summarized in Table 4.
6. DISCUSSION
In this article, we surveyed representative and
state-of-the-art methods related to information dif-
fusion analysis in online social networks, ranging
from popular topic detection to diusion modeling
techniques, including methods for identifying inu-
ential spreaders. Figure 8 presents the taxonomy of
the various approaches employed to address these
issues. Hereafter we provide a discussion regarding
their shortcomings and related open problems.
6.1 Detecting Popular Topics
The detection of popular topics from the stream
of messages produced by the members of an OSN re-
lies on the identication of bursts. There are mainly
two ways to detect such patterns, by analyzing (i)
term frequency or (ii) social interaction frequency.
In this area, the following challenges certainly need
to be addressed:
Topic denition and scalability. It is obvi-
ous that not all methods dene a topic in the same
way. For instance Peaky Topics simply assimilates
a topic to a word. It has the advantage to be a low
complexity solution, however, the produced result is
of little interest. In contrast, OLDA denes a topic
as a distribution over a set of words but in turn has
a high complexity, which prevents it from being ap-
plied at large scale. Consequently, there is a need
for new methods that could produce intelligible re-
sults while preserving eciency. We identify two
possible ways to do so, through: (i) the conception
of new scalable algorithms, or (ii) improved imple-
mentations of the algorithms using, e.g. distributed
systems (such as Hadoop).
Social dimension. Furthermore, popular topic
detection could be improved by leveraging bursti-
ness and people authority, as does TSTE, which
relies on the PageRank algorithm. However, that
possibility remains ill explored so far.
Data complexity. Currently the focus is set on
the textual content exchanged in social networks.
However, more and more often, users exchange other
types of data such as images, videos, URLs point-
ing to those objects or Web pages, etc. This situa-
tion has to be fully considered and integrated at the
heart of the eorts carried out to provide a complete
solution for topic detection.
6.2 Modeling Information Diffusion
We distinguish two types of models, explanatory
and predictive. Concerning predictive models, on
the one hand there are non-graph based methods,
that are limited by the fact that they ignore the
topology of the network and only forecast the evo-
lution of the rate at which information globally dif-
fuses. On the other hand, there are graph based
approaches that are able to predict who will inu-
ence whom. However, they cannot be used when
the network is unknown or implicit. Although a
lot of eort have been performed in this area, gen-
erally speaking, there is a need to consider more
realistic constraints when studying information dif-
fusion. In particular, the following issues have to be
dealt with:
DEFINITION 10 (Closed World). The
closed world assumption holds that information can
only propagate from node to node via the network
edges and that nodes cannot be inuenced by exter-
nal sources.
Closed world assumption. The major obser-
vation about modeling information diusion is cer-
tainly that all the described approaches work under
a closed world assumption, dened in Denition 10.
In other words, they assume that people can only
be inuenced by other members of the network and
that information spreads because of informational
cascades. However, most observed spreading pro-
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
Information
diffusion in onIine
sociaI networks
Detecting
interesting topics
ModeIing diffusion
processes
Identifying
infIuentiaI
spreaders
Bursty and
emergent
topics
ExpIanatory
modeIs
Predictive
modeIs
TopoIogicaI
approaches
Other
approaches
Using term
frequency
Using sociaI
interaction
frequency
Static network
Dynamic
network
Graph-based
Non graph-
based
Using diffusion
modeIs
Using users'
features
C
H
A
L
L
E
N
G
E
S
I
S
S
U
E
A
P
P
R
O
A
C
H
E
S
Incorporating
opinion
detection
Taking topic
into account
Taking competing and
cooperating information
into account
Defining
topics more
preciseIy
Improving
scaIabiIity
Incorporating
sociaI
properties
Taking topic
into account
-- AREAS FOR IMPROVEMENT --
ReIaxing the
cIosed worId
assumption
Figure 8: The above taxonomy presents the three main research challenges arising from in-
formation diusion in online social networks and the related types of approaches, annotated
with areas for improvement.
cesses in OSNs do not rely solely on social inu-
ence. The closed-world assumption is proven incor-
rect in recent work on Twitter done by Myers et
al. [32] in which authors observe that information
tends to jump across the network. The study shows
that only 71% of the information volume in Twit-
ter is due to internal inuence and the remaining
29% can be attributed to external events and inu-
ence. Consequently they provide a model capable
of quantifying the level of external exposure and in-
uence using hazard functions [10]. To relax this
assumption, one way would be to align users pro-
les across multiple social networking sites. In this
way, it would be possible to observe the information
diusion among various platforms simultaneously
(subject to the availability of data). Some work
tend to address this type of problems by proposing
to de-anonymize the social networks [33].
Cooperating and competing diusion pro-
cesses. In addition, the described studies rely on
the assumption that diusion processes are inde-
pendent, i.e. each information spreads in isolation.
Myers et al. [31] argue that spreading processes
cooperate and compete. Competing contagions de-
crease each others probability of diusion, while
cooperating ones help each other in being adopted.
They propose a model that quanties how dierent
spreading cascades interact with each other. It pre-
dicts diusion probabilities that are on average 71%
more or less than the diusion probability would be
for a purely independent diusion process. We be-
lieve that models have to consider and incorporate
this knowledge.
Topic-sensitive modeling. Furthermore, it is
important for predictive models to be topic-sensitive.
Romero et al. [39] have studied Twitter and found
signicant dierences in the mechanics of informa-
tion diusion across topics. More particularly, they
have observed that information dealing with politi-
cally controversial topics are particularly persistent,
with repeated exposures continuing to have unusu-
ally large marginal eects on adoption, which val-
idates the complex contagion principle that stipu-
lates that repeated exposures to an idea are par-
ticularly crucial when the idea is controversial or
contentious.
Dynamic networks. Finally, it is important
to note that OSNs are highly dynamic structures.
Nonetheless most of the existing work rely on the as-
sumption that the network remains static over time.
Integrating link prediction could be a basis to im-
prove prediction accuracy. A more complete review
of literature on this topic can be found in [20].
6.3 Identifying Inuential Spreaders
There are various ways to tackle this issue, rang-
ing from pure topological approaches, such as k-
shell decomposition or HITS to textual clustering
based approaches, including hybrid methods, such
as IP which combines the HITS algorithm with
nodes features. As mentioned previously, there is
no such thing as a universal inuencer and therefore
topic-sensitive methods have also been developed.
Opinion detection. The notion of inuence is
strongly linked to the notion of opinion. Numer-
ous studies on this issue have emerged in recent
years, aiming at automatically detecting opinions
or sentiment from corpus of data. We believe that
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
it might be interesting to include this kind of work
in the context of information diusion. Work deal-
ing with the diusion of opinions themselves have
emerged [29] and it seems that there is an interest
to couple these approaches.
6.4 Applications
Even if there are a lot of contributions in the
domain of online social networks dynamics analy-
sis, we can remark that implementations are rarely
provided for re-use. What is more, available imple-
mentations require dierent formatting of the in-
put data and are written using various program-
ming languages, which makes it hard to evaluate or
compare existing techniques. SONDY [18] intends
to facilitate the implementation and distribution of
techniques for online social networks data mining.
It is an open-source tool that provides data pre-
processing functionalities and implements some of
the methods reviewed in this paper for topic de-
tection and inuential spreaders identication. It
features a user-friendly interface and proposes visu-
alizations for topic trends and network structure.
7. REFERENCES
[1] L. AlSumait, D. Barbara, and C. Domeniconi.
On-line lda: Adaptive topic models for mining
text streams with applications to topic
detection and tracking. In ICDM 08, pages
312, 2008.
[2] A. Anagnostopoulos, R. Kumar, and
M. Mahdian. Inuence and correlation in
social networks. In KDD 08, pages 715,
2008.
[3] E. Bakshy, I. Rosenn, C. Marlow, and
L. Adamic. The role of social networks in
information diusion. In WWW 12, pages
519528, 2012.
[4] D. Blei, A. Ng, and M. Jordan. Latent
dirichlet allocation. The Journal of Machine
Learning Research, 3:9931022, 2003.
[5] P. Brown and J. Feng. Measuring user
inuence on Twitter using modied k-shell
decomposition. In ICWSM 11 Workshops,
pages 1823, 2011.
[6] M. Cataldi, L. Di Caro, and C. Schifanella.
Emerging topic detection on Twitter based on
temporal and social terms evaluation. In
MDMKDD 10, pages 413, 2010.
[7] M. D. Choudhury, Y.-R. Lin, H. Sundaram,
K. S. Candan, L. Xie, and A. Kelliher. How
does the data sampling strategy impact the
discovery of information diusion in social
media? In ICWSM 10, pages 3441, 2010.
[8] T. F. Coleman and Y. Li. A reective newton
method for minimizing a quadratic function
subject to bounds on some of the variables.
SIAM J. on Optimization, 6(4):10401058,
Apr. 1996.
[9] I. CVX Research. CVX: Matlab software for
disciplined convex programming, version 2.0
beta. http://cvxr.com/cvx, sep 2012.
[10] R. C. Elandt-Johnson and N. L. Johnson.
Survival Models and Data Analysis. John
Wiley and Sons, 1980/1999.
[11] W. Galuba, K. Aberer, D. Chakraborty,
Z. Despotovic, and W. Kellerer. Outtweeting
the twitterers - predicting information
cascades in microblogs. In WOSN 10, pages
311, 2010.
[12] C. F. Gerald and P. O. Wheatley. Applied
numerical analysis with MAPLE; 7th ed.
Addison-Wesley, Reading, MA, 2004.
[13] J. Goldenberg, B. Libai, and E. Muller. Talk
of the network: A complex systems look at
the underlying process of word-of-mouth.
Marketing Letters, 2001.
[14] M. Gomez-Rodriguez, D. Balduzzi, and
B. Scholkopf. Uncovering the temporal
dynamics of diusion networks. In ICML 11,
pages 561568, 2011.
[15] M. Gomez Rodriguez, J. Leskovec, and
A. Krause. Inferring networks of diusion and
inuence. In KDD 10, pages 10191028, 2010.
[16] M. Gomez-Rodriguez, J. Leskovec, and
B. Schokopf. Structure and dynamics of
information pathways in online media. In
WSDM 13, pages 2332, 2013.
[17] M. Granovetter. Threshold models of
collective behavior. American journal of
sociology, pages 14201443, 1978.
[18] A. Guille, C. Favre, H. Hacid, and D. Zighed.
Sondy: An open source platform for social
dynamics mining and analysis. In
SIGMOD 13, (demonstration) 2013.
[19] A. Guille and H. Hacid. A predictive model
for the temporal dynamics of information
diusion in online social networks. In
WWW 12 Companion, pages 11451152,
2012.
[20] M. A. Hasan and M. J. Zaki. A survey of link
prediction in social networks. In Social
Network Data Analytics, pages 243275.
Springer, 2011.
[21] H. W. Hethcote. The mathematics of
infectious diseases. SIAM REVIEW,
42(4):599653, 2000.
[22] P. N. Howard and A. Duy. Opening closed
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3
regimes, what was the role of social media
during the arab spring? Project on
Information Technology and Political Islam,
pages 130, 2011.
[23] A. Hughes and L. Palen. Twitter adoption
and use in mass convergence and emergency
events. International Journal of Emergency
Management, 6(3):248260, 2009.
[24] D. Kempe. Maximizing the spread of inuence
through a social network. In KDD 03, pages
137146, 2003.
[25] M. Kitsak, L. Gallos, S. Havlin, F. Liljeros,
L. Muchnik, H. Stanley, and H. Makse.
Identication of inuential spreaders in
complex networks. Nature Physics,
6(11):888893, Aug 2010.
[26] J. Kleinberg. Bursty and hierarchical
structure in streams. In KDD 02, pages
91101, 2002.
[27] J. Leskovec, L. Backstrom, and J. Kleinberg.
Meme-tracking and the dynamics of the news
cycle. In KDD 09, pages 497506, 2009.
[28] J. Leskovec, M. Mcglohon, C. Faloutsos,
N. Glance, and M. Hurst. Cascading behavior
in large blog graphs. In SDM 07, pages
551556, (short paper) 2007.
[29] L. Li, A. Scaglione, A. Swami, and Q. Zhao.
Phase transition in opinion diusion in social
networks. In ICASSP 12, pages 30733076,
2012.
[30] J. Makkonen, H. Ahonen-Myka, and
M. Salmenkivi. Simple semantics in topic
detection and tracking. Inf. Retr.,
7(3-4):347368, Sept. 2004.
[31] S. Myers and J. Leskovec. Clash of the
contagions: Cooperation and competition in
information diusion. In ICDM 12, pages
539548, 2012.
[32] S. A. Myers, C. Zhu, and J. Leskovec.
Information diusion and external inuence in
networks. In KDD 12, pages 3341, 2012.
[33] A. Narayanan and V. Shmatikov.
De-anonymizing social networks. In SP 09,
pages 173187, 2009.
[34] M. E. J. Newman. The structure and function
of complex networks. SIAM Review,
45:167256, 2003.
[35] L. Page, S. Brin, R. Motwani, and
T. Winograd. The pagerank citation ranking:
Bringing order to the web. In WWW 98,
pages 161172, 1998.
[36] A. Pal and S. Counts. Identifying topical
authorities in microblogs. In WSDM 11,
pages 4554, 2011.
[37] E. M. Rogers. Diusion of Innovations, 5th
Edition. Free Press, 5th edition, aug 2003.
[38] D. Romero, W. Galuba, S. Asur, and
B. Huberman. Inuence and passivity in
social media. In ECML/PKDD 11, pages
1833, 2011.
[39] D. M. Romero, B. Meeder, and J. Kleinberg.
Dierences in the mechanics of information
diusion across topics: idioms, political
hashtags, and complex contagion on Twitter.
In WWW 11, pages 695704, 2011.
[40] L. Rong and Y. Qing. Trends analysis of news
topics on Twitter. International Journal of
Machine Learning and Computing,
2(3):327332, 2012.
[41] E. Sadikov, M. Medina, J. Leskovec, and
H. Garcia-Molina. Correcting for missing data
in information cascades. In WSDM 11, pages
5564, 2011.
[42] K. Saito, K. Ohara, Y. Yamagishi,
M. Kimura, and H. Motoda. Learning
diusion probability based on node attributes
in social networks. In ISMIS 11, pages
153162, 2011.
[43] G. Salton and C. Buckley. Term-weighting
approaches in automatic text retrieval. Inf.
Process. Manage., 24(5):513523, 1988.
[44] G. Salton and M. J. McGill. Introduction to
Modern Information Retrieval. McGraw-Hill,
1986.
[45] S. B. Seidman. Network structure and
minimum degree. Social Networks, 5(3):269
287, 1983.
[46] D. A. Shamma, L. Kennedy, and E. F.
Churchill. Peaks and persistence: modeling
the shape of microblog conversations. In
CSCW 11, pages 355358, (short paper)
2011.
[47] T. Takahashi, R. Tomioka, and K. Yamanishi.
Discovering emerging topics in social streams
via link anomaly detection. In ICDM 11,
pages 12301235, 2011.
[48] F. Wang, H. Wang, and K. Xu. Diusive
logistic model towards predicting information
diusion in online social networks. In
ICDCS 12 Workshops, pages 133139, 2012.
[49] J. Weng, E.-P. Lim, J. Jiang, and Q. He.
TwitterRank: nding topic-sensitive
inuential twitterers. In WSDM 10, pages
261270, 2010.
[50] J. Yang and J. Leskovec. Modeling
information diusion in implicit networks. In
ICDM 10, pages 599608, 2010.
h
a
l
-
0
0
8
4
8
0
5
0
,

v
e
r
s
i
o
n

1

-

2
5

J
u
l

2
0
1
3