The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Socio-semantic dynamics in a blog network
Jean-Philippe Cointet CREA (CNRS/EP, France)
AND
Camille Roth CAMS (CNRS/EHESS, France)
IEEE S OCIAL C OM 09, VANCOUVER , BC – AUG 29–31, 2009
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
A SOCIAL network Three kinds of links for each blog... citation: post citation links
citation link
interaction: comment links affiliation: blogroll links ...where contents circulate in terms of topics (W) in terms of cultural items (U)
Dataset: US blogosphere scope: 4 months of ’08 campaign network: citations nodes: 1, 066 blogs (RTGI)
comment link
blogroll link
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
A socio-SEMANTIC network Three kinds of links for each blog... citation: post citation links interaction: comment links affiliation: blogroll links ...where contents circulate in terms of topics (W) in terms of cultural items (U)
Dataset: US blogosphere
semantic characterization
scope: 4 months of ’08 campaign
“relevant” syntagms
network: citations
(“health insurance”, “climate change”, “national
nodes: 1, 066 blogs (RTGI)
security”, “super Tuesday”, “human rights”...)
urls:
“www.youtube.com/x1hqwkeac”, etc.
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
A DYNAMIC socio-semantic network Three kinds of links for each blog... citation: post citation links interaction: comment links affiliation: blogroll links ...where contents circulate in terms of topics (W) in terms of cultural items (U)
Dataset: US blogosphere scope: 4 months of ’08 campaign network: citations nodes: 1, 066 blogs (RTGI)
http://presidentialwatch08.com/
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Socio-semantic configuration 0.06 super tuesday michigan california huckabee
frequency of terms
0.05
0.04
0.03
0.02
0.01
0
Jan 1
Michigan Primary
time
a
20/02 20/02 19/02 c
b
Super Tuesday
26/02
d
feb 17
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Socio-semantic configuration
semantic profile of a blog i:
0.2 0.18
Wi (w) ˆ i (w) := W P|W| Wi (w) w=1
0.14
|B|
|{j, Wj (w) > 0}|
0.12
P (δ)
· log
0.16
0.1 0.08 0.06 0.04 0.02
semantic distance between blogs i and j: δ(i, j) = 1 −
0
[0;.1[ [.1;.2[ [.2;.3[ [.3;.4[ [.4;.5[ [.5;.6[ [.6;.7[ [.7;.8[ [.8;.9[ [.9;1]
δ
ˆ i ·W ˆj W ˆ ˆ j" "Wi ""W
Semantic distance distributions. Triangles: computed over the whole set of possible blog pairs. Crosses: distribution computed on linked blogs.
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Computing link creation propensity
→ estimate the “propensity of interaction” ...that it is more or less likely for a node (or a dyad) with property “m” to receive a link ...which may be simply estimated by: ˆf (m) = ν(m) N(m) ν(m) = number of links pointing towards an agent of type m
(resp. number of new dyads of type m) during a time period,
N(m) = number of agents (resp. of dyads) of type m.
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Computing link creation propensity
ˆf (m) = ν(m) N(m)
1
10
fˆ(k)
→ estimate the “propensity of interaction” ...that it is more or less likely for a node (or a dyad) with property “m” to receive a link ...which may be simply estimated by:
0
10
0
50
100
k
150
ν(m) = number of links pointing towards an agent of type m
(resp. number of new dyads of type m) during a time period,
N(m) = number of agents (resp. of dyads) of type m.
200
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing topological distance effects
→ strong trend to repetition and local semantic distance
1
10
fˆ(k)
interaction
0
10
→ strong trend to homophily 0
primarily “social”? social distance and degree
50
100
k
150
200
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
2
10
topological distance effects interaction
ˆ f (d)
→ strong trend to repetition and local
1
10
0
10
semantic distance
→ strong trend to homophily primarily “social”? social distance and degree
−1
10
−2
10
1
2
3
d
4
>4
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
2
10
topological distance effects interaction
semantic distance
1
10
ˆ g(δ)
→ strong trend to repetition and local
0
10
→ strong trend to homophily primarily “social”? social distance and degree
−1
10
[0;.1] ].1;.2] ].2;.3] ].3;.4] ].4;.5] ].5;.6] ].6;.7] ].7;.8] ].8;.9] ].9;1]
δ
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing 0
10
topological distance effects interaction
semantic distance
→ strong trend to homophily primarily “social”? social distance and degree
−1
propension p(d, δ)
→ strong trend to repetition and local
10
−2
10
−3
10
−4
10
1 2 3 social distance d
>3
[0;0.2[
[0.8;1] [0.2;0.4[ [0.4;0.6[ [0.6;0.8[ semantic distance δ
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing 2
topological distance effects interaction
semantic distance
→ strong trend to homophily
propension p(d, k)
→ strong trend to repetition and local
10
1
10
0
10
−1
10
−2
10
1 2 3
primarily “social”? social distance and degree
>3 social distance d
0
50
100
social capital k
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Information flows: measures on the post network a
Dyadic measures: raw, weighted network, aggregated on 4 months
1
b
2
1 2
attentional matrix a... → and total attention αa = 5/6 detachment matrix
1
3
c
2
3 1 1 f 2
“edge range”: quantifying shortcuts
1 d
3
e
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Information flows: measures on the post network a
Dyadic measures: raw, weighted network, aggregated on 4 months
1/6 3/4 b
attentional matrix a... → and total attention αa = 5/6
1/4
2/3
1/3 2/6 c
1/5
detachment matrix 3/6
2/2 1/5 f 2/3
“edge range”: quantifying shortcuts
1/3 d
3/5
e
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Information flows: measures on the post network a
Dyadic measures:
6 4/3
raw, weighted network, aggregated on 4 months
b
4
3/2
3 3
attentional matrix a... → and total attention αa = 5/6
c
5
detachment matrix 2
1 5 f 3/2
“edge range”: quantifying shortcuts
3 d
5/3
e
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Information flows: measures on the post network a
6
Dyadic measures: raw, weighted network, aggregated on 4 months
4/3
4
3/2
b 3
attentional matrix a... → and total attention αa = 5/6
c 1
f
detachment matrix 2
3/2
5
3
5 e
“edge range”: quantifying shortcuts d
5/3
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
Information cascade Diffusion subgraphs 4
10
a
20/02
b
26/02
d
20/02 19/02 c
An example of diffusion subgraph, a common “resource” and a set of citation links between blogs
number of diffusion subgraphs
links nodes
3
10
2
10
1
10
0
10 0 10
1
10
size of diffusion subgraphs
⇒ heterogeneous cascade sizes
2
10
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
An ego-centered perspective
role of the total attention on the number of diffusion links
2
a bit more “global”...
Number of tranmissions
10
We focus on the total number of “transmissions” generated by blogs with a given total attention α
1
10
0
10
−1
10
−2
10
second transmissions: we focus on “later transmissions”, i.e. after a first transmission event
−1
10
0
10
Total Attention α
1
10
2
10
Larger active readership => larger number of diffusion links, yet not linearly
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
An ego-centered perspective
role of the total attention on the number of diffusion links
2
a bit more “global”...
Number of tranmissions
10
We focus on the total number of “transmissions” generated by blogs with a given total attention α
1
10
0
10
−1
10
−2
10
−1
10
0
10
Total Attention α
1
10
2
10
second transmissions: we focus on “later transmissions”, i.e. after a first transmission event
Larger active readership => larger number of diffusion links, yet not linearly
The “blogosphere” as a socio-semantic network
Link creation dynamics
Diffusion dynamics
→ role of edge-range on the number of grand-children We focus again on transmissions occurring after a first transmission event
a
20/02 20/02 19/02 c
b
26/02
d
Mean number of 2n d transmissions
A more global perspective 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0
10
20
30
40
50
60
edge range r
70
80
An information which has been transmitted through a “median” link generates a larger number of grandchildren
90
100
The “blogosphere” as a socio-semantic network
Link creation dynamics
Concluding remarks
Co-evolution of content and relationships Patterns not necessarily linked to authority only Patterns not necessarily ego-centered only → divergent from the “neighbor-based-influence” perspective
Diffusion dynamics
The “blogosphere” as a socio-semantic network
Link creation dynamics
Concluding remarks
Co-evolution of content and relationships Patterns not necessarily linked to authority only Patterns not necessarily ego-centered only → divergent from the “neighbor-based-influence” perspective Thank you!
[email protected] &
[email protected]
Diffusion dynamics
