The document analyzes face-to-face behavioral networks using wearable RFID devices to study human interactions in social environments, focusing on two case studies: the Science Gallery in Dublin and the HT09 conference in Turin. It discusses the network structure, resilience, and the dynamics of information spreading through human interactions, while emphasizing the importance of understanding network properties for applications in computer science and epidemiology. The findings highlight the relationship between network topology and the spread of information, illustrating the variability of dynamics across different social settings.

1. What’s in a crowd? Analysis of face-to-face
behavioral networks
Lorenzo Isella1 , Alain Barrat1,2 , Juliette Stehlé2 ,
Jean-François Pinton3 , Wouter Van den Broeck1 and
Ciro Cattuto1
1 Complex Networks and Systems Group, Institute for Scientific Interchange (ISI)
Foundation, Turin, Italy.
2 Centre de Physique Théorique, CNRS UMR 6207, Marseille, France.
3 Laboratoire de Physique de l’ENS Lyon, CNRS UMR 5672, Lyon, France.
Workshop on data driven dynamical networks, Les
Houches, France, 2010

2. Outline
Overview of the RFID infrastructure deployed to mine for
face-to-face proximity ⇒ networks of human interactions.
Network structural analysis.
Network resilence.
Information spreading: longitudinal network ⇐⇒ causality
reachability and variability
kinetics of information spreading
Conclusions.

3. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

4. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

5. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

6. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

7. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

8. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

9. Goals and Case Studies
Deployment of wearable RFID devices to collect data
about human interaction in different social environments.
Focus on two main case studies
Science Gallery (SG) at the Trinity College, Dublin, Ireland
(∼ 3 months, ∼ 10000 visitors).
HT09 conference, Turin, Italy (3 days, ∼ 100 participants).
Patented technology and data from Sociopatterns Project
(http://www.sociopatterns.org/).
Applications in computer science (ubiquitous computing,
P2P) and computational epidemiology (causality,
non-homogeneous mixing).

10. Overview of the Infrastructure
Tags exchange packets at various powers and report their
contacts to antennas broadcasting the data to a server.
Low-power packets expose face-to-face interactions at
small distances (∼ 1m).

11. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

12. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

13. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

14. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

15. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

16. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

17. From Physical Proximity to Networks
Natural representation of physical proximity as a network in
addition to
scalability
unobtrusiveness
low cost
high spatial resolution ∼ 1 meter
high temporal resolution ∼ 5 − 20 seconds.

18. Aggregated Networks
Aggregate all the contacts along 24 hours.
HT09: June, 30th SG: July, 14th
q q
q
q
q
q q
q
q
q q
q
q q q
q
SG: May, 19th SG: May, 20th
q q
q
q
q
q
q
q q
q
q q
q
q
q q
q q
q
q q
q q q

19. Human Dynamics and Network Topology 1/2
Entanglement of human behavior and network topology.
0.008
P (visit duration)
0.006
0.004
0.002
0.000
101 102
Visit duration (min)
12:00 to 13:00
13:00 to 14:00
14:00 to 15:00
15:00 to 16:00
16:00 to 17:00
17:00 to 18:00
18:00 to 19:00
19:00 to 20:00

20. Human Dynamics and Network Topology 2/2
Short-tailed P(k ) and broad P(wij ) and P(∆tij ).
SG HT09
10-1 10-1
10-2
10-2
P (k)
P (k)
-3
10
10-3
10-4
10-5 10-4
0 10 20 30 40 50 60 70 0 20 40 60 80
k k
100 100
-1
10 SG
10-1 SG
HT09 HT09
10-2 -2
10
P (∆tij )
P (wij )
10-3
10-3
10-4
10-5 10-4
-6
10 10-5
101 102 103 104 101 102 103 104
∆tij (sec) wij (sec)

21. Random Networks and Smallworldness
Network topology↔ information spreading.
HT09: June, 30th (rewired) SG: July, 14th (rewired)
q
q
q
q
q
q
q
q
q
q
q
HT09: June, 30th SG: July, 14th
1.0 1.0
0.8
0.8
M(l)/M(∞)
M(l)/M(∞)
0.6
0.6
0.4
0.4
Aggregated network 0.2 Aggregated network
Rewired networks Rewired networks
0.2 0.0
1 2 3 4 1 2 3 4 5 6 7 8 9 10
l l

22. Dismantling strategies 1/2
Removal strategies expose network structure.
Cumulative duration and/or sophisticated measures
(Onnela et al., PNAS,104, 7332 (2007)), similarity, etc..
Oij = 0 Oij = 1/3
i j i j
Oij = 2/3 Oij = 1
i j i j

23. Dismantling strategies 2/2
Topology-based strategies enhance network
fragmentation.
Removing strong links as least effective strategy.
HT2009: June, 30th Dublin: July, 14th
300
100
250
80
200
60
150
N1
N1
increasing wij increasing wij
40 decreasing wij decreasing wij
increasing Oij 100 increasing Oij
increasing simij increasing simij
20 50
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Removal Fraction Removal Fraction

24. Deterministic SI model 1/2
SI model S + I → 2I, infection probability .
Set = 1: snowball deterministic model (avoid
stochasticity).
Beyond epidemiology: paradigm for information diffusion
and causality on the network.
I S I I
+

25. Deterministic SI model 1/2
SI model S + I → 2I, infection probability .
Set = 1: snowball deterministic model (avoid
stochasticity).
Beyond epidemiology: paradigm for information diffusion
and causality on the network.
I S I I
+
Collect distributions of infected visitors/conference
participants at the end of each day by varying the seed
(inter day variability).

26. Deterministic SI model 1/2
SI model S + I → 2I, infection probability .
Set = 1: snowball deterministic model (avoid
stochasticity).
Beyond epidemiology: paradigm for information diffusion
and causality on the network.
I S I I
+
Collect distributions of infected visitors/conference
participants at the end of each day by varying the seed
(inter day variability).
Dependence of the epidemic spreading during a single day
on the choice of the seed (intra day variability).

27. Deterministic SI model 2/2
Processes of and on the network
partially aggregated network [human contacts]
transmission network [information spreading].
transmission network ⊆ partially aggregated network
nodes outside seed’s CC cannot be reached by infection
Fastest path = shortest path.

28. Deterministic SI model 2/2
Processes of and on the network
partially aggregated network [human contacts]
transmission network [information spreading].
transmission network ⊆ partially aggregated network
nodes outside seed’s CC cannot be reached by infection
Fastest path = shortest path.

29. Deterministic SI model 2/2
Processes of and on the network
partially aggregated network [human contacts]
transmission network [information spreading].
transmission network ⊆ partially aggregated network
nodes outside seed’s CC cannot be reached by infection
Fastest path = shortest path.

30. Inter day variability
Nsus for a given seed ≡ number of individuals in the seed’s
CC.
Ninf
In a static network framework, P(Ninf /Nsus ) = δ( Nsus − 1).
Information propagates differently at HT09 and SG.
HT09 SG
1.0 1.0
0.8 0.8
P (Ninf /Nsus )
P (Ninf /Nsus )
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Ninf /Nsus Ninf /Nsus

31. Intra day variability
Impact of social events (e.g. coffee breaks).
Highlight role played by each seed (hard to achieve in a
static network framework).
HT09: June, 30th SG: July, 14th
300
100 8:00 to 9:00 12:00 to 13:00
9:00 to 10:00 13:00 to 14:00
10:00 to 11:00 250 14:00 to 15:00
80 11:00 to 12:00 15:00 to 16:00
Incidence curve
Incidence curve
12:00 to 13:00 16:00 to 17:00
13:00 to 14:00 200 17:00 to 18:00
14:00 to 15:00 18:00 to 19:00
60 15:00 to 16:00 19:00 to 20:00
16:00 to 17:00 150
40
100
20 50
0 0
08:00 10:00 12:00 14:00 16:00 18:00 20:00 12:00 14:00 16:00 18:00 20:00
Time Time

32. Kinetics of information spreading 1/2
Examples from collected data at HT09
Network diameters going back and forth in time.
q qqq
q qq q q
q q q q q
qqq qq
q qqqq q q qq
q q q q
q q q q qqqqqqqqqqqqq
qqq q q q q q qqq qq
q q q q q q q
q qq qq qq q q qqqqq
qqqqqqqqqqqqqqqq qq qqqq
q q q q q q q q q
qqqqq qq q q q q q q qqq
qqqq qq q q q qqqqq
q q q q q q
q q q qqq qq
q q

33. Kinetics of information spreading 2/2
Distribution of shortest vs fastest path length.
SG: May, 19th SG: May, 20th
0.3 0.4
Transmission network Transmission network
Aggregated network 0.3 Aggregated network
0.2
P (nd )
P (nd )
0.2
0.1
0.1
0.0 0.0
1 2 3 4 5 6 7 8 9 10 1 3 5 7 9 11 13
nd nd
SG: July, 14th HT09: June, 30th
0.3 0.7
0.6
Transmission network Transmission network
Aggregated network 0.5 Aggregated network
0.2
0.4
P (nd )
P (nd )
0.3
0.1
0.2
0.1
0.0 0.0
1 3 5 7 9 11 13 15 17 19 1 3 5 7 9 11
nd nd

34. Conclusions
Network aggregation over different time periods
(seasonality, trends, etc..).
Network static properties (P(k ), diameter, assortativity,
etc..).
Network dynamic properties: spread of epidemics and
diffusion processes on a longitudinal network hence
dynamics of the network and dynamics on the network.
What’s in a crowd? Analysis of face-to-face behavioral
networks, L.Isella et al.
http://arxiv.org/abs/1006.1260.

35. Conclusions
Network aggregation over different time periods
(seasonality, trends, etc..).
Network static properties (P(k ), diameter, assortativity,
etc..).
Network dynamic properties: spread of epidemics and
diffusion processes on a longitudinal network hence
dynamics of the network and dynamics on the network.
What’s in a crowd? Analysis of face-to-face behavioral
networks, L.Isella et al.
http://arxiv.org/abs/1006.1260.

36. Conclusions
Network aggregation over different time periods
(seasonality, trends, etc..).
Network static properties (P(k ), diameter, assortativity,
etc..).
Network dynamic properties: spread of epidemics and
diffusion processes on a longitudinal network hence
dynamics of the network and dynamics on the network.
What’s in a crowd? Analysis of face-to-face behavioral
networks, L.Isella et al.
http://arxiv.org/abs/1006.1260.

37. Conclusions
Network aggregation over different time periods
(seasonality, trends, etc..).
Network static properties (P(k ), diameter, assortativity,
etc..).
Network dynamic properties: spread of epidemics and
diffusion processes on a longitudinal network hence
dynamics of the network and dynamics on the network.
What’s in a crowd? Analysis of face-to-face behavioral
networks, L.Isella et al.
http://arxiv.org/abs/1006.1260.

38. Conclusions
Network aggregation over different time periods
(seasonality, trends, etc..).
Network static properties (P(k ), diameter, assortativity,
etc..).
Network dynamic properties: spread of epidemics and
diffusion processes on a longitudinal network hence
dynamics of the network and dynamics on the network.
What’s in a crowd? Analysis of face-to-face behavioral
networks, L.Isella et al.
http://arxiv.org/abs/1006.1260.

39. Conclusions
Network aggregation over different time periods
(seasonality, trends, etc..).
Network static properties (P(k ), diameter, assortativity,
etc..).
Network dynamic properties: spread of epidemics and
diffusion processes on a longitudinal network hence
dynamics of the network and dynamics on the network.
What’s in a crowd? Analysis of face-to-face behavioral
networks, L.Isella et al.
http://arxiv.org/abs/1006.1260.

40. Acknowlegements
Michael John Gorman, director of the Science Gallery at
Trinity College, Dublin
http://sciencegallery.com/content/
science-gallery-2009-infectious
Organizers of Hypertext 2009 conference
http://www.ht2009.org/
SocioPatterns project and partners
http://www.sociopatterns.org
DynaNets project
http://www.dynanets.org/
Thank you for your attention!

41. Acknowlegements
Michael John Gorman, director of the Science Gallery at
Trinity College, Dublin
http://sciencegallery.com/content/
science-gallery-2009-infectious
Organizers of Hypertext 2009 conference
http://www.ht2009.org/
SocioPatterns project and partners
http://www.sociopatterns.org
DynaNets project
http://www.dynanets.org/
Thank you for your attention!

42. Acknowlegements
Michael John Gorman, director of the Science Gallery at
Trinity College, Dublin
http://sciencegallery.com/content/
science-gallery-2009-infectious
Organizers of Hypertext 2009 conference
http://www.ht2009.org/
SocioPatterns project and partners
http://www.sociopatterns.org
DynaNets project
http://www.dynanets.org/
Thank you for your attention!

43. Acknowlegements
Michael John Gorman, director of the Science Gallery at
Trinity College, Dublin
http://sciencegallery.com/content/
science-gallery-2009-infectious
Organizers of Hypertext 2009 conference
http://www.ht2009.org/
SocioPatterns project and partners
http://www.sociopatterns.org
DynaNets project
http://www.dynanets.org/
Thank you for your attention!

44. Acknowlegements
Michael John Gorman, director of the Science Gallery at
Trinity College, Dublin
http://sciencegallery.com/content/
science-gallery-2009-infectious
Organizers of Hypertext 2009 conference
http://www.ht2009.org/
SocioPatterns project and partners
http://www.sociopatterns.org
DynaNets project
http://www.dynanets.org/
Thank you for your attention!

45. Acknowlegements
Michael John Gorman, director of the Science Gallery at
Trinity College, Dublin
http://sciencegallery.com/content/
science-gallery-2009-infectious
Organizers of Hypertext 2009 conference
http://www.ht2009.org/
SocioPatterns project and partners
http://www.sociopatterns.org
DynaNets project
http://www.dynanets.org/
Thank you for your attention!