Activity driven modeling of time varying

networks
SUBJECT AREAS:
N. Perra1,2, B. Gonçalves1, R. Pastor-Satorras3 & A. Vespignani1,4,5
STATISTICAL PHYSICS,
THERMODYNAMICS AND
NONLINEAR DYNAMICS 1
Department of Physics, College of Computer and Information Sciences, Department of Health Sciences, Northeastern University,
THEORETICAL PHYSICS Boston MA 02115 USA, 2Linkalab, Cagliari, Italy, 3Departament de Fı́sica i Enginyeria Nuclear, Universitat Politècnica de
MODELLING AND THEORY Catalunya, Campus Nord B4, 08034 Barcelona, Spain, 4Institute for Scientific Interchange Foundation, Turin 10133, Italy, 5Institute
for Quantitative Social Sciences at Harvard University, Cambridge, MA, 02138.
GENERAL PHYSICS

Network modeling plays a critical role in identifying statistical regularities and structural principles
Received common to many systems. The large majority of recent modeling approaches are connectivity driven. The
28 March 2012 structural patterns of the network are at the basis of the mechanisms ruling the network formation.
Connectivity driven models necessarily provide a time-aggregated representation that may fail to describe
Accepted the instantaneous and fluctuating dynamics of many networks. We address this challenge by defining the
7 June 2012 activity potential, a time invariant function characterizing the agents’ interactions and constructing an
activity driven model capable of encoding the instantaneous time description of the network dynamics. The
Published
model provides an explanation of structural features such as the presence of hubs, which simply originate
25 June 2012 from the heterogeneous activity of agents. Within this framework, highly dynamical networks can be
described analytically, allowing a quantitative discussion of the biases induced by the time-aggregated
representations in the analysis of dynamical processes.
Correspondence and

N
requests for materials etwork modeling1–6 has long drawn on the tradition of social network analysis and graph theory, with
should be addressed to models ranging from the Erdös-Rényi model to Logit models, p*-models, and Markov random graphs7–11.
N.P. (n.perra@neu. In the last decade, the class of growing network models, exemplified by the preferential attachment model,
edu) has been made widely popular by research in statistical physics and computer science12–17. All these models can be
defined as connectivity driven, as the network’s topology is at the core of the model’s algorithmic definition.
Connectivity-driven network models are well-suited for capturing the essential features of systems such as the
Internet, where connections among nodes are long-lived elements18,19. However, in many cases the interactions
among the elements of the system are rapidly changing and are characterized by processes whose timing and
duration are defined on a very short time scale20,21. This limit has been investigated in the case of adaptive systems
whose structure evolve being coupled to the process taking place on top of them22–26. Instead, the understating of
this limit in time varying networks in which the structure evolves independently of the process is still limited and
unexplored. In these activity-driven networks, models intended to capture the process of accumulating connec-
tions over time and the resulting degree distribution (i.e. the probability that a node has k connections to other
nodes) and other topological properties merely represent a time-integrated perspective of the system.
Furthermore, the analysis of dynamical processes in evolving networks is generally performed in the presence
of a time-scale separation between the network evolution and the dynamical process unfolding on its structure. In
one limit we can consider the network as quenched in its connectivity pattern, thus evolving on a time scale that is
much longer that the dynamical process itself. In the other limiting case, the network is evolving at a time scale
much shorter than the dynamical process thus effectively disappearing from the definition of the interaction
among individuals that is conveniently replaced by effective random couplings. While time scale separation is
extremely convenient for the numerical and analytical tractability of the models, networks generally evolve on a
time-scale that might be comparable to the one of the dynamical process27–32. An accurate modelization of the
dynamics of activity-driven networks calls for the definition of interaction processes based on
the actual measurement of the activity of the agents forming the system, a task now made possible by the
availability of time-resolved, high-quality data on the instantaneous activity of millions of agents in a wide variety
of networks33–37.

Results
Here we present the analysis of three large-scale, time-resolved network datasets and define for each node
a measurable quantity, the activity potential, characterizing its interaction pattern within the network. This

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 1

 www.nature.com/scientificreports

measure is defined as the number of interactions performed, in a Activity driven network model. Our empirical analysis naturally
given time window, by each node divided by the total number of leads to the definition of a simple model that uses the activity
interactions made by all the nodes in the same time window. We find distribution to drive the formation of a dynamic network. We
that the system level dynamics of the network can be encoded by the consider N nodes (agents) and assign to each node i an activity/
activity potential distribution function from which it is possible to firing rate ai 5 gxi, defined as the probability per unit time to
derive the appropriate interaction rate among nodes. On the basis of create new contacts or interactions with other individuals, where g
the empirically measured activity potential distribution we propose a is a rescaling factor defined such that the average number of active
process model for the generation of random dynamic networks. The nodes per unit time in the system is gÆxæN. The activity rates are
activity potential function defines the network structure in time and defined such that the numbers xi are bounded in the interval
traces back the origin of hubs to the heterogenous activity of the ƒxi ƒ1, and are assigned according to a given probability
network elements. The model allows to write dynamical equations distribution F(x) that may be chosen arbitrarily or given by
coupling the network dynamics and the dynamical processes unfold- empirical data. We impose a lower cut-off on x in order to avoid
ing on its structure without relying on any time-scale separation possible divergences of F(x) close to the origin. We assume a simple
approximation. We analyze a simple spreading process and provide generative process according to the following rules (see Fig. 2-D):
the explicit analytical expression for the biases introduced by the
time-aggregated representation of the network when studying
. At each discrete time step t the network Gt starts with N dis-
dynamical processes occurring on a time scale comparable to that connected vertices;
of the network evolution. Interestingly, the network model presented . With probability aiDt each vertex i becomes active and generates
here is amenable to the introduction of many features in the nodes’ m links that are connected to m other randomly selected vertices.
dynamic such as the the persistency of specific interactions or assort- Non-active nodes can still receive connections from other active
ative/disassortative correlations, thus defining a general basic mod- vertices;
eling framework for rapidly evolving networks. . At the next time step t 1 Dt, all the edges in the network Gt are
deleted. From this definition it follows that all interactions have a
The activity potential. We consider three datasets corresponding to constant duration ti 5 Dt.
networks in which we can measure the individual agents’ activity: The above model is random and Markovian in the sense that
Collaborations in the journal ‘‘Physical Review Letters’’ (PRL) agents do not have memory of the previous time steps. The full
published by the American Physical Society38, messages exchanged dynamics of the network and its ensuing structure is thus completely
over the Twitter microblogging network, and the activity of actors encoded in the activity potential distribution F(x).
in movies and TV series as recorded in the Internet Movie Database In Fig. 3 we report the results of numerical simulations of a net-
(IMDb)39. In the first dataset the network representation con- work with N 5 5000, m 5 2, g 5 10, and F(x) / x2c, with c 5 2.8 and
siders undirected links connecting two PRL authors if they have ~10{3 . The model recovers the same qualitative behavior observed
collaborated in writing one article. In the second system each node in Fig. 1. At each time step the network is a simple random graph with
is a Twitter user and an undirected link is drawn if at least one low average connectivity. The accumulation of connections that we
message has been exchanged between two users. Finally, the actor observe by measuring activity on increasingly larger time slices T
network is obtained by drawing an undirected link between any two generates a skewed PT(k) degree distribution with a broad variability.
actors who have participated in the same movie or TV series. The presence of heterogeneities and hubs (nodes with a large number
Simple evidence for the role of agents’ activity in network analysis of connections) is due to the wide variation of activity rates in the
and modeling can be readily observed in the case of the collaboration system and the associated highly active agents. However, it is worth
network of scientific authors40. The number of collaborations of any remarking that hub formation has a different interpretation than in
author depends on the time window through which we observe the growing network prescriptions, such as preferential attachment. In
system. In Fig. 1 we show the networks obtained by time-aggregated those cases hubs are created by a positional advantage in degree space
co-authorships over 1, 10, and 30 years for the subset of authors in the leading to the passive attraction of more and more connections. In
PRL dataset who were active in the considered time period. Clearly, our model, the creation of hubs results from the presence of nodes
the time scale used to construct the network defines a non-stationary with high activity rate, which are more willing to repeatedly engage in
connectivity pattern and explicitly affects the network structure and its interactions.
degree distribution. Similar results are found for the other two datasets The model allows for a simple analytical treatment. We define the
as shown in the Supplementary Information. S
integrated network GT ~ t~T t~0 Gt as the union of all the networks
In the three datasets considered, we characterize the individual obtained in each previous time step. The instantaneous network
activity of every agent: papers written, messages exchanged, or movie generated at each time t will be composed of a set of slightly inter-
appearances, respectively. For each dataset we measure the individual connected nodes corresponding to the agents that were active at that
activity of each agent and define the activity potential xi of the agent i particular time, plus those who received connections from active
as the number of interactions that he/she performs in a characteristic agents. Each active node will create m links and the total edges per
time window of given length Dt, divided by the total number of unit time are Et 5 mNgÆxæ yielding the average degree per unit time
interactions made by all agents during the same time window. The the contact rate of the network
activity potential xi thus estimates the probability that the agent i was 2Et
involved in any given interaction in the system, and the probability hk it ~ ~2mghxi: ð1Þ
distribution F(x) that a randomly chosen agent i has activity potential N
x statistically defines the interaction dynamics of the system. In Fig. 2 The instantaneous network will be composed by a set of stars, the
we show the cumulative distribution Fc(x) evaluated for the three vertices that were active at that time step, with degree larger than or
datasets. In all cases we find that, contrary to the degree distribution equal to m, plus some vertices with low degree. The corresponding
and other structural characteristics of the networks, the distribution integrated network, on the other hand, will generally not be sparse,
Fc(x) is virtually independent of the time scale over which the activity being the union of all the instantaneous networks at previous times
potential is measured. Additionally, we find that the distribution Fc(x) (see Fig. 3). In fact, for large time T and network size N, when the
is skewed and fairly broadly distributed. This is hardly surprising as in degree in the integrated network can be approximated by a continu-
many cases measurements of human activity have confirmed the ous variable, we can show (see Supplementary Information) that
presence of wide variability across individuals41,42. agent i will have at time T a degree in the integrated network given

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 2

 www.nature.com/scientificreports

Figure 1 | Network visualization and degree distribution of the PRL dataset considering three different aggregated views. In particular, in the first two
rows we focus on the set of authors who wrote at least one paper in the period between 1960 and 1974. For this subset of 5, 162 active authors we construct
three different networks, graphically represented in the central row of the figure. The upper row represents a blown up perspective of a particular network
region. In the left column we show the network of 1974, defined by the active nodes in the given time frame. The central column shows the network
obtained by integrating over 10 years, from 1974 to 1984. In the right column we show the network obtained by integrating over 30 years, from 1974
to2004. The first network is highly fragmented as is obvious from the visualization. When larger windows are integrated the density of the network
increases and heterogeneous connectivity patterns start to emerge. Clearly, as indicated by the degree distributions, that consider the complete set of
authors (not just those used for the sake of visualization in the first two rows), the time scale used to construct the network affects its topological structure.
In each visualization the size and color of the nodes is proportional to their degree.

 
by ki ðT Þ~N 1{e{Tmgxi =N . It can then easily be shown that the explored connections are more likely to happen again), social rela-
degree distribution PT(k) of the integrated network at time T takes tions have a lifetime distribution (persistence) and multiple connec-
the form: tions and weighted links may be relevant. Neither of these effects is
 
k considered in the model. We report some statistical analysis of those
PT ðkÞ*F , ð2Þ features in the Supplementary Information as further ingredients to
Tmg
be considered in future extensions of the model.
where we have considered the limit of small k/N and k/T (i.e. large
network size and times). The noticeable result here is the relation Dynamical processes in activity driven networks. Recent research
between the degree distribution of the integrated network and the has highlighted the key role of interaction dynamics as opposed to
distribution of individual activity, which, from the previous equa- static studies. For example, an individual who appears to be central
tion, share the same functional form. This relation is approximately by traditional network metrics may in fact be the last to be infected
recovered in the empirical data, where the activity potential distri- because of the timing of his/her interactions30,43. Analogously the
bution is in reasonable agreement with the appropriately rescaled concurrency of sexual partners can dramatically accelerate the
asymptotic degree distribution of the corresponding network (see spread of STDs31. Despite its simplicity, our model makes it
Fig. 4-A). As expected, differences between the two distributions analytically explicit that the actors’ activity time scale plays a major
are present, due to features of the real network dynamics that our role in the understanding of processes unfolding on dynamical
random model does not capture: links might have memory (already networks. Let us consider the susceptible-infected-susceptible (SIS)

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 3

 www.nature.com/scientificreports

Figure 2 | Cumulative distribution of the activity potential, FC(x), empirically measured by using four different time windows and a schematic
representation of the proposed network model. In particular, in panel (A) we show the cumulative distributions of the observables x for Twitter, in panel
(B) for IMDb, and in panel (C) for PRL. In panel (D) we show a schematic representation of the model. Considering just 13 nodes and m 5 3, we plot a
visualization of the resulting networks for 3 different time steps. The red nodes represent the firing/active nodes. The final visualization represents the
network after integration over all time steps.

epidemic compartmental model1,2,44,45. In this model, infected instantaneous networks whose union defines the integrated
individuals can propagate the disease to healthy neighbors with one30,31,43,51. In Fig. 4-B we plot the results of numerical simulations
probability l, while infected individuals recover with rate m and of the SIS model on a network generated according to our model and
become susceptible again. In an homogenous population the on two time-aggregated network instances. We observe that the two
behavior of the epidemics is controlled by the reproductive aggregated networks lead to misleading results in both the threshold
number R0 5 b/m, where b 5 lÆkæ is the per capita spreading rate and the epidemic magnitude as a function of b/m. Even if the
that takes into account the rate of contacts of each individual. The epidemic threshold discounts the different average degree of the
reproductive number identifies the average number of secondary networks in the factor b 5 lÆkæ, the two aggregated instances
cases generated by a primary case in an entirely susceptible consider all edges as always available to carry the contagion
population and defines the epidemic threshold such that only if R0 process, disregarding the fact that the edges may be active or not
. 1 can epidemics reach an endemic state and spread into a closed according to a specific time sequence defined by the agents’ activity.
population. In the past few years the inclusion of complex The above finding can be more precisely quantified by calculating
connectivity networks and mobility schemes into the substrate of analytically the epidemic threshold in activity driven networks with-
spreading processes contagion, diffusion, transfer, etc. has out relying on any time aggregated view of the network connectivity.
highlighted new and interesting results46–50. Several results states By working with activity rates we can derive epidemic evolution
that the epidemic threshold depends on the topological properties equation in which the spreading process and the network dynamics
of the networks. In particular, for networks characterized by a fix, are coupled together. Let us assume a distribution of activity poten-
quenched topology the threshold is given by the principal eigenvalue tial x of nodes given by a general distribution F(x) as before. At a
of the adjacency matrix48,49. Instead, for annealed network, cha- mean-field level, the epidemic process will be characterized by the
racterized by a topology defined just on average because the number of infected individuals in the class of activity rate a, at time t,
connectivity patterns has a dynamic extremely fast with respect to namely Iat . The number of infected individuals of class a at time t 1
the dynamical process, heterogeneous mean-field approaches2,6 Dt given by:
predict an epidemic threshold that is inversely proportional to the IatzDt ~{mDtIat zIat z
second moment of the network’s degree distribution: b/m . Ækæ2/Æk2æ. ð ð
However, these results do not apply to the case in which the time   It   I t a’Dt
lm Nat {Iat aDt da’ a’ zlm Nat {Iat da’ a’ , ð3Þ
variation of the connectivity pattern is occurring on the same time N N
scale of the dynamical process. Our model presents simple evidence
of this problem, as a disease with a small value of m21 (the infectious where Na is the total number of individuals with activity a. In Eq. (3),
period characteristic time) will have time to explore the fully- the third term on the right side takes into account the probability that
integrated network, but will not spread on the dynamic a susceptible of class a is active and acquires the infection getting a

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 4

 www.nature.com/scientificreports

Figure 3 | Visualization and degree distributions of the proposed network model considering different aggregated views. We fix N 5 5000, m 5 2, g 5
10, F(x) / x2c with c 5 2.8, # x # 1 with 5 1023. We plot the network obtained after one time step in the first column, the network obtained after
integrating over 10 iterations in the second column, and the network obtained after integrating over 20 iterations in the last column. Interestingly, even
though the model is random and markovian by construction, we observe a behavior qualitatively similar to the case of PRL: the single time window yields
a sparse and poorly connected network with a trivial degree distribution. When larger time scales are considered, heterogeneous connectivity patterns
start to emerge as seen by the corresponding degree distributions. In each visualization the size and color of the nodes is proportional to their degree.

connection from any other infected individual (summing over all for which longitudinal data are available. This function allows the
different classes), while the last term takes into account the probabil- definition of a simple dynamical process based on the nodes’ activity
ity that a susceptible, independently of his activity, gets a connection rate, providing a time dependent description of the network’s con-
from any infected active individual. The above equation can be nectivity pattern. Despite its simplicity, the model can be used to
solved as shown in the material and methods section, yielding the solve analytically the co-evolution of the network and contagion
following epidemic threshold for the activity driven model: processes and characterize quantitatively the biases generated by
b 2hai time-scale separation techniques. Furthermore the proposed model
w pffiffiffiffiffiffiffiffi : ð4Þ appears to be suited as a testbed to discuss the effect of network
m haiz ha2 i
dynamics on other processes such as damage resilience, discovery
This result considers the activity rate of each actor and therefore and data mining, collective behavior and synchronization. While we
takes into account the actual dynamics of interactions; the above have reduced the level of realism for the sake of parsimony of the
formula does not depend on the time-aggregated network repres- presented model, we are aware of the importance of analyzing other
entation and provides the epidemic threshold as a function of the features of actor activity such as concurrency, persistence and differ-
interaction rate of the nodes. This allows to characterize the spread- ent weights associated with each connection. These features must
ing condition on the natural time scale of the combination of the necessarily be added to the model in order to remove the limitations
network and spreading process evolution. set by the simple random network structures generated here and
represent interesting challenges for future work in this area.
Discussion
We have presented a model of dynamical networks that encodes the Methods
connectivity pattern in a single function, the activity potential dis- Datasets. We considered three different dataset: the collaborations in the journal
tribution, that can be empirically measured in real world networks ‘‘Physical Review Letters’’ (PRL) published by the APS, the message exchanged on

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 5

 www.nature.com/scientificreports

ð
daIatzDt ~I tzDt ~I t {mDtI t zlmhaiDtI t zlmht Dt, ð5Þ
Ð
where ht ~ da’Ia’t a’ and we have dropped all second order terms in the activity rate a
t
and in Ia . We are not considering events in which two infected nodes choose each
other for connection and we are considering a linear approximation in Iat since in the
beginning of the epidemics the number of infectious individuals in each class is small.
In order to obtain an closed expression for h we multiply both sides of Eq. (3) by a and
integrate over all activity spectrum, obtaining the equation
 
htzDt ~ht {mDtht zlm a2 I t Dtzlmhaiht Dt: ð6Þ

In the continuous time limit we obtain the following closed system of equations
Lt I~{mIzlmhaiIzlmh, ð7Þ
 
Lt h~{mhzlm a2 Izlmhaih, ð8Þ
whose Jacobian matrix has eigenvalues
pffiffiffiffiffiffiffiffi
Lð1,2Þ ~hailm{m+lm ha 2 i: ð9Þ

The epidemic threshold for the system is obtained requiring the largest eigenvalues to
be larger the 0, which leads to the condition for the presence of an endemic state:
l 1 1
w pffiffiffiffiffiffiffiffi ð10Þ
m m haiz ha2 i

From this last expression we can recover the epidemic threshold of Eq. (4) by con-
sidering b 5 lÆkæ, ai 5 gxi and Ækæ 5 2mgÆxæ.

1. Newman, M. Networks. An Introduction (Oxford Univesity Press, 2010).

2. Barrat, A., Barthélemy, M. & Vespignani, A. Dynamical Processes on Complex
Figure 4 | In panel (A) we consider the entire Twitter dataset and show Networks (Cambridge Univesity Press, 2008).
the distribution of activity potential F(x) and the asymptotic degree 3. Albert, R. & Barabási, A.-L. Statistical mechanics of complex networks. Rev. Mod.
distribution of the corresponding network, PT[jk], with j 5 1/(Tgm), Phys. 74, 47 (2002).
4. Boccaletti, S., Latora, V., Moreno, Y., Chavez, M. & Hwang, D.-U. Complex
rescaled according to the analytical result. In panel (B) we show the
networks: Structure and dynamics. Physics Reports 424, 175 (2006).
density of infected nodes, I‘, in the stationary state, obtained from 5. Bollobas, B. Moder Graph Theory (Springer-Verlag, 1998).
numerical simulations of the SIS model on a network generated according 6. Vespignani, A. Modeling dynamical processes in complex socio-technical
to the proposed model and two other networks resulting from an systems. Nature Physics 8, 32–39 (2012).
integration of the model over 20 and 40 time steps, respectively. We set N 7. Erdös, P. & Rényi, A. On random graphs. Publications mathematicae 6, 290
(1959).
5 106, m 5 5, g 5 10, F(x) / x2c with c 5 2.1 and # x # 1 with 5 1023.
8. Molloy, M. & Reed, B. A critical point for random graphs with a given degree
Each point represents an average over 102 independent simulations. sequence. Random Structures and Algorithms 6, 161 (1995).
The red triangle marks the critical reproductive number Rco as predicted by 9. Holland, P. & Leinhardt, S. An exponential family of probabilty distributions of
Eq. 4. directed graphs. J. Am. Stat. Assoc. 76, 33–50 (1981).
10. Frank, O. & Strauss, D. Markov graphs. J. Am. Stat. Assoc. 81, 832–842 (1986).
11. Wasserman, S. & Pattiston, P. Logit models and logistic regression for social
Twitter and the activity of actors in movies and TV series as recorded in the Internet networks. Psychometrika 61, 401–425 (1996).
Movie Database (IMDb). In particular: 12. Barabási, A.-L., Albert, R. & Jeong, H. Mean-field theory for scale-free random
networks. Physica A 272, 173 (1999).
PRL dataset. In this database the network representation considers each author of a 13. Barabási, A.-L. & Albert, R. Emergence of scaling in random networks. Science
PRL article as a node. An undirected link between two different authors is drawn if 286, 509 (1999).
they collaborated in the same article. We filter out all the articles with more than 10 14. Dorogovtsev, S. N., Mendes, J. & Samukhin, A. N. Structure of growing networks
authors in order to focus our attention just on small collaborations in which we can with preferential linking. Phys. Rev. Lett. 85, 4633 (2000).
assume that the social components is relevant. We consider the period between 1960 15. Dorogovtsev, S. N. & Mendes, J. F. F. Evolution of Networks: From Biological nets
and 2004. In this time window we registered 71, 583 active nodes and 261, 553 to the Internet and WWW (2003).
connections among them. In this dataset is natural defining the activity rate, a, of each 16. Fortunato, S., Flammini, A. & Menczer, F. Scale-free network growth by ranking.
author as the number of papers written in a specific time window Dt 5 1 year. Phys. Rev. Lett. 96, 218701 (2006).
Authors with no collaborative papers in the total time span considered (isolates) are 17. Boguña, M. & Pastor-Satorras, R. Class of correlated random networks with
not included in the data set. hidden variables. Phys. Rev. E 68, 036112 (2011).
18. Pastor-Satorras, R. & Vespignani, A. Evolution and Structure of the Internet : A
Twitter Dataset. Having been granted temporary access to Twitter’s firehose we Statistical Physics Approach (Cambridge University Press, 2004).
mined the stream for over 6 months to identify a large sample of active user accounts. 19. Albert, R., Jeong, H. & Barabási, A.-L. The diameter of the world wide web. Nature
Using the API, we then queried for the complete history of 3 million users, resulting in 401, 130–131 (1999).
a total of over 380 million individual tweets covering almost 4 years of user activity on 20. Holme, P. & Saramäki, J. Temporal networks. arxiv:1108.1780 (2011).
Twitter. In this database the network representation considers each users as a node. 21. Ghoshal, G. & Holme, P. Attractiveness and activity in internet communities.
An undirected link between two different users is drawn if they exchanged at least one Physica A 364, 603–609 (2005).
message. We focus our attention on 9 months during 2008. In this time window we 22. Volz, E. & Meyers, L. A. Epidemic thresholds in dynamic contact networks.
registered 531, 788 active nodes and 2, 566, 398 connections among them. In this J. R. Soc. Interface 6, 233241 (2009).
dataset we define the activity rate of each user as the number of messages sent in a time 23. Centola, D., Gonzalez-Avella, J. C., Eguiluz, V. M. & San Miguel, M. Homophily,
window Dt 5 1 day. cultural drift, and the co-evolution of cultural groups. J. Conflict Resolution 51,
905929 (2007).
IMDb Dataset. In this database the network representation considers each actor as a 24. Jolad, S., Liu, W., Schmittmann, B. & Zia, R. K. P. Epidemic spreading on preferred
node. An undirected link between two different actors is drawn if they collaborated in degree adaptive networks. arxiv:1109.5440 (2011).
the same movie/TV series. We focus on the period between 1950 and 2010. During 25. Schwartz, I. & Shaw, L. Rewiring for adaptation. Physics 3, 17 (2010).
this time period we registered 1, 273, 631 active nodes and 47, 884, 882 connections 26. Shaw, L. B. & Schwartz, I. B. Enhanced vaccine control of epidemics in adaptive
between them. A natural way to define the activity rate in this dataset is to consider the networks. Phys. Rev. E 81, 046120 (2010).
number of movies acted by each actor in a specific time window Dt 5 1 year. 27. Butts, C. Revisting the foundations of network analysis. Science 325, 414–416
(2009).
Epidemic threshold. In order to solve Eq. (3) we can consider the total number of 28. Butts, C. Relational event framework for social action. Sociological Methodology
infectious nodes in the system 38, 155–200 (2008).

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 6

 www.nature.com/scientificreports

29. Panisson, A. et al. On the dynamics of human proximity for data diffusion in ad- 48. Wang, Y., Chakrabarti, D., Wang, C. & Faloutsos, C. Epidemic spreading in real
hoc networks. Ad Hoc Networks (2011). networks: An eigenvalue viewpoint. 22nd Symposium on Reliable Distributed
30. Moody, J. The importance of relationship timing for diffusion: Indirect Computing (SRDS2003) (2003).
connectivity and std infection risk. Soc. Forces 81, 25 (2002). 49. Chakrabarti, D., Wang, Y., Wang, C., Leskovec, J. & Faloutsos, C. Epidemic
31. Morris, M. & Kretzschmar, M. Concurrent partnerships and the spread of hiv. thresholds in real networks. ACM Transactions on Information and System
AIDS 11, 641 (1997). Security (TISSEC), 10 (4) (2008).
32. Basu, P., Guha, S., Swami, A. & Towsley, D. Percolation phenomena in networks 50. Castellano, C. & Pastor-Satorras, R. Thresholds for epidemic spreading in
under random dynamics. In COMSNETS 2012, Bangalore, India (2012). networks. Phys. Rev. Lett. 105, 218701 (2010).
33. González, M. C., Hidalgo, C. A. & Barabási, A.-L. Understanding individual 51. Morris, M. Sexually Transmitted Diseases, K.K. Holmes, et al. Eds. (McGraw-Hill,
human mobility patterns. Nature 453, 779 (2008). 2007).
34. Onnela, J.-P. et al. Structure and tie strengths in mobile communication networks.
Proc. Natl. Acad. Sci. U.S.A. 104, 7332 (2007).
35. Lazer, D. et al. Computational social science. Science 323, 721 (2009).
36. Vespignani, A. Predicting the behavior of techno-social systems. Science 325, 425– Acknowledgments
428 (2009). The work has been partly sponsored by the Army Research Laboratory and was
accomplished under Cooperative Agreement Number W911NF-09-2-0053. RPS
37. Brockmann, D., Hufnagel, L. & Geisel, T. The scaling laws of human travel. Nature
acknowledges financial support from the Spanish MICINN (project
439, 462 (2006).
FIS2010-21781-C02-01) and additional support through ICREA Academia, funded by the
38. APS. Data sets for research (2010).
Generalitat de Catalunya.
39. IMDb. Internet movie database http://www.imdb.com/interfaces (2010). Date of
access 10/11/2011.
40. Newman, M. E. J. The structure of scientific collaboration networks. Proc. Natl. Author contributions
Acad. Sci. 98, 404 (2001). R.P.-S. & A.V designed research, N.P. performed simulations, N.P., B.G, R.P.-S. & A.V.
41. Barabási, A.-L. The origin of bursts and heavy tails in human dynamics. Nature analyzed the data, N.P., R.P.-S. & A.V. contributed new analytical results. All authors wrote,
435, 207 (2005). reviewed and approved the manuscript.
42. Jo, H.-H., Karsai, M., Kertész, J. & Kaski, K. Circadian pattern and burstiness in
human communication activity. New. J. Phys 14, 013055 (2012).
43. Isella, L. et al. What’s in a crowd? analysis of face-to-face behavioral networks. Additional information
Supplementary information accompanies this paper at http://www.nature.com/
J. Theor. Biol. 271, 166–180 (2011).
scientificreports
44. Kermack, W. O. & McKendrick, A. G. A contribution to the mathematical theory
of epidemics. Proc. R. Soc. A 115, 700 (1927). Competing financial interests: The authors declare no competing financial interests.
45. Keeling, M. & Rohani, P. Modeling Infectious Disease in Humans and Animals License: This work is licensed under a Creative Commons
(Princeton University Press, 2008). Attribution-NonCommercial-ShareAlike 3.0 Unported License. To view a copy of this
46. Lloyd, A. & May, R. How viruses spread among computers and people. Science license, visit http://creativecommons.org/licenses/by-nc-sa/3.0/
292, 1316 (2001). How to cite this article: Perra, N., Gonçalves, B., Pastor-Satorras, R. & Vespignani, A.
47. Balcan, D. et al. Multiscale mobility networks and the spatial spreading of Activity driven modeling of time varying networks. Sci. Rep. 2, 469; DOI:10.1038/srep00469
infectious diseases. Proc. Natl. Acad. Sci. U.S.A. 106, 21484 (2009). (2012).

SCIENTIFIC REPORTS | 2 : 469 | DOI: 10.1038/srep00469 7

 Reproduced with permission of copyright owner.
Further reproduction prohibited without permission.