Post on 01-Jul-2015
description
What Stops Social Epidemics?
Greg Ver Steeg
Rumi Ghosh & Kris:na Lerman
USC Informa:on Sciences Ins:tute
2
Informa:on, viruses, etc. spread from node to node on a network
Transmissibility, λ = Probability to infect your neighbor
Infected
Not Infected
• What is an epidemic? We observe many cascades that: – Grow quickly ini:ally – But remain too small for standard (viral) epidemic models
• Informa:on cascades differ: – Response to repeated exposure is important on Digg (and TwiVer)
– Dras:cally alters predic:ons about size of epidemics
3
What is an epidemic?
4
Frac:on of nodes infected
1
Transmissibility, λ
Epidemic threshold predicted for many cascade models 0
On an infinite graph, an epidemic is any process that spreads to a frac:on of all the nodes
Social news:
Distribu:on of cascade size on -‐-‐-‐-‐-‐-‐
6
#nodes ∼300k
Most cascades less than 1% of total network size!
A small frac:on is s:ll a frac:on, though, right?
Why are these cascades so small?
7
Standard model of epidemic growth
(Heterogenous mean field theory, SIR model, same degree distribu:on as Digg)
Transmissibility of almost all Digg stories fall within width of this line?!
λ, Transmissibility
Most cascades fall in this range
Maybe graph structure is responsible?
clustering reduces epidemic threshold and cascade size, but not enough!
transmissibility λ
epidemic threshold
← Mean field predic:on (same degree dist.)
← Simulated cascades on a random graph with same degree dist.
Simulated cascades on the observed Digg graph
8
What about the spreading mechanism?
9
Independent Cascade Model implicit in many epidemic models
Infected
Not Infected
?
How important are repeat exposures?
10
More than half exposed to a story more than once!
How do people respond to repeated exposure?
11
Not much.
We have similar results for TwiVer -‐-‐-‐-‐-‐-‐-‐
Also noted by Romero, et al, WWW 2011
Big consequences for epidemic growth
• Most people are exposed to a story more than once
• Repeated exposures have liVle effect
• Growth of epidemics is severely curtailed (especially compared to Ind. Cascade Model)
12
Weak response to repeated exposure
13
Take effect of repeat exposure into account:
Actual Digg cascades
Result of simula:ons
λ*
Epidemic threshold unchanged
λ*, Transmissibility
Also explains dynamics
14
Number of new people exposed to a story (who don’t vote on it)
Number of new people exposed to a story (who do vote)
Transmissibility: the percentage of new people exposed who end up infected/vo:ng
15 Approximate :me of story promo:on to front page
Structure + Behavior
16
Accurate model
of behavior
Independent cascade model
=+
+ =
+ =
absolutelytrue
msaleem
upickjaybol
noupsell
anderzole
vtbarrera
badwithcomputer
xdvx
Bukowsky
1KrazyKorean
kevinrose
skored
IvanBlouiebaur
AmyVernon
oboy
Burento
MrBabyMan TalSiach
absolutelytrue
msaleem
upickjaybol
noupsell
anderzole
vtbarrera
badwithcomputer
xdvx
Bukowsky
1KrazyKorean
kevinrose
skored
IvanBlouiebaur
AmyVernon
oboy
Burento
MrBabyMan TalSiach
Accurate model
of behavior
Summary
• Informa:on spread ≠ Disease spread • Big consequences for epidemics
• Repeat exposures are important on Digg and TwiVer
• On Digg, people don’t respond to repeat exposure – Epidemic threshold unchanged
– Dras:cally reduces size of epidemics
17
Decay of novelty
18
Weak response to repeated exposure
19
Standard model of epidemic growth
(Heterogenous mean field theory, SIR model, same degree distribu:on as Digg)
λ*
What is an epidemic on a finite graph?
20
Infected
Not Infected
We would call this an epidemic, right?
And now?
Epidemics saturate the graph
What is an epidemic on a finite graph?
21
Same number of red dots
Epidemics saturate the graph
Sub-‐epidemic cascade size
22
Reproduc:ve number, R0 is the average number of new spreaders reached by each spreader.
R0 < 1 ⇒ No Epidemic
R0 = �k�λ
Average number of friends
Transmissibility
1�average degree0
50
100
150
200
Λ, transmissibility
Cascadesize
23
Transmissibility of almost all Digg stories fall within width of this line?!
Sub-epidemic cascade size = 1 +1
1−R0
Satura:on on a real graph
24
25
26