sgb presentation
DESCRIPTION
Together with colleagues from the University of Zaragoza, Dr. Gonzalez-Bailon applied advanced mathematical methods to the analysis of the role played by Twitter in facilitating protest mobilization in Spain in 2011. Their findings “shed light on the connection between online networks, social contagion, and collective dynamics, and offer an empirical test to the recruitment mechanisms theorized in formal models of collective action.”TRANSCRIPT
3/15/2012
1
T i ’ ‘b R l iTweetin’ ‘bout a Revolution
The growth of the Spanish Occupy movementin an online network
Joint work with
Javier Borge-Holthoefer and Yamir Moreno
Person of the Year 2011: The Protester• The word ‘protest’ appears inThe word protest appears in newspapers exponentially more in 2011 than ever before
• Global tipping point for frustration
• SNSs did not cause the movements,
Dec 26, 2011
but kept them alive
• Technology helped spread “the virus of protest”
3/15/2012
2
New Media and Social Unrest
Wave of Protests (2011): Timeline of Events
01 02 03 04 05 06 07 08 09 10 11 12
January
Tunisia Algeria Saudi Arabia
February
Yemen Bahrain Libya
May
Spain GreeceChile
July
Israel
August
UK
September
US
December
Russia
Egypt Syria Jordan
y
3/15/2012
3
Research Question
The Spatial Diffusion of Protests
Q
Overlapping Networks
Research QuestionQ
3/15/2012
4
Networks and Collective Action
Some historical examples:
• insurgency in Paris commune in1871 (Gould 1991)
• the 60s civil right struggles in the USthe 60s civil right struggles in the US (McAdam 1986)
• demonstrations in East Germany (Opp and Gern 1993; Lohman 1994)
Threshold Models
Watts, D. J., & Dodds, P. S. (2010). Threshold Models of Social Influence. In P. Bearman & P. Hedström (Eds.), Handbook of Analytical Sociology, OUP
3/15/2012
5
Threshold Models – Main Findings
• the shape of threshold distribution determines the global reach ofthe shape of threshold distribution determines the global reach ofcascades;
• individual thresholds interact with the size of local networks;
• critical mass depends on activating large number of low threshold actors that are well connected in the overall structure;
• exposure to multiple sources can be more important than multiple exposures from the same source (complex contagion)
Influence in Online Networks
Centola, D. (2010). “The Spread of Behavior in an Online Social Network Experiment”. Science
Romero, D. M., Meeder, B., & Kleinberg, J. (2011). “Differences in the Mechanics of Information Diffusion Across Topics”, WWW
3/15/2012
6
Activism in the Internet Age
• lower costs outdates resource mobilisation theory
• what is ‘success’ in the era flash activism?activism?
The Spanish ‘Indignados’ Movement
3/15/2012
7
Data
#hashtags
15m.bifi.es
3/15/2012
8
The Twitter Network: StatisticsFull SymmetricalFull
NetworkSymmetrical
Network
N (# nodes) 87,569 80,715M (# arcs) 6,030,459 2,644,367<k> (avg degree) 69 33C (clustering) 0.220 0.198l (path length) 3.24 3.65D (diameter) 11 11( )r (assortativity) -0.139 -0.0344# strong components 5,249 139N giant component 82,253 80,421N 2nd component 4 4max(kin) (# following) 5,773 5,082max(kout) (# followers) 31,798 5,082
Distribution of Users in the Network by Activitycounts
0
1
2
3
ssa
ge
s re
ceiv
ed
/se
nt
15000
7500
1 influentials
2hidden
influentials
103
102
101
0
ssa
ge
s re
ceiv
ed
/se
nt
-4 -3 -2 -1 0 1 2
-2
-1
ratio following/followers
ratio
me
s
1
3 4broadcasters
common users
10‐4 10‐3 10‐2 10‐1 0 101 102
10‐1
10‐2
ratio
me
s
3/15/2012
9
The Online Growth of the Movement
Recruitment and Activation Thresholds
• Activation time: moment when users start emitting protest messagesActivation time: moment when users start emitting protest messages
• ka/kin ≈ 0 low threshold individuals (no need of ‘local pressure’)
• ka/kin ≈ 1 high threshold individuals (need high ‘local pressure’)
ka/kin = 2/4 = 0.5 ka/kin = 4/8 = 0.5
3/15/2012
10
Distribution of Thresholds
Offline Media Coverage of Protests
News Search – keywords used: "15-M" or "indignados" or "democracia real"
3/15/2012
11
Distribution of Thresholds before and after 15-M
Threshold Distribution by Activation Day
3/15/2012
12
Percentage of Activated Users by Type
Information Cascades
RTs ‘Spike Trains’RTs Spike Trains
(Bakshy et al. 2011) t t + Δt t + 2Δt
3/15/2012
13
Information Cascades
Where are Recruiters and Spreaders?k-shell decompositionp
(Kitsak et al. 2010)
3/15/2012
14
Where are the Recruiters?
Where are the Spreaders?
3/15/2012
15
Where are the Spreaders?
influentials hidden influentials
A. activation timeinfluentials hidden influentialsinfluentials hidden influentials
B. cascade size
-30000 -20000 -10000 0
0.00
000
0.00
010
-30000 -20000 -10000 015‐M 15‐M
broadcasters common users0
12
34
5
casc
ad
e s
ize
(log
)
01
23
45
casc
ad
e s
ize
(log
)
broadcasters common usersbroadcasters common users
-30000 -20000 -10000 0-30000 -20000 -10000 0
0.00
000
0.00
010
15‐M15‐M
broadcasters common users0
12
34
5
casc
ad
e s
ize
(log
)
01
23
45
casc
ad
e s
ize
(log
)
Summary of Findings
• feedback between dynamics of recruitment and information diffusion
• being central is crucial for diffusion, not so for recruitment
• exogenous factors create random seeding in the network
3/15/2012
16
Discussion
To main limitations:
• we do not control for homophily
• we do not control for exposure to offline media
so we might be overestimating influence
So Gladwell got it wrong…