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, V ANCOUVER , BC – A UG 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 comment link blogroll link nodes : 1 , 066 blogs (RTGI)
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 (“health insurance”, “climate change”, “national network : citations security”, “super Tuesday”, “human rights”...) nodes : 1 , 066 blogs (RTGI) 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 http://presidentialwatch08.com/ network : citations nodes : 1 , 066 blogs (RTGI)
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics Socio-semantic configuration 0.06 super tuesday michigan california 0.05 huckabee frequency of terms 0.04 0.03 0.02 0.01 0 Jan 1 Michigan Primary Super Tuesday feb 17 time a 20/02 b 26/02 d 20/02 19/02 c
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 0.16 W i ( w ) ˆ W i ( w ) := 0.14 P |W| w = 1 W i ( w ) 0.12 |B| P ( δ ) 0.1 · log |{ j , W j ( w ) > 0 }| 0.08 0.06 0.04 0.02 0 semantic distance [0;.1[ [.1;.2[ [.2;.3[ [.3;.4[ [.4;.5[ [.5;.6[ [.6;.7[ [.7;.8[ [.8;.9[ [.9;1] δ between blogs i and j : W i · ˆ ˆ W j Semantic distance distributions. Triangles: computed over the δ ( i , j ) = 1 − whole set of possible blog pairs. Crosses: distribution computed on � ˆ W i �� ˆ W j � 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: f ( m ) = ν ( m ) ˆ 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 → 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 1 10 ...which may be simply estimated by: f ( k ) ˆ f ( m ) = ν ( m ) 0 10 ˆ N ( m ) 0 50 100 150 200 k 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 Dynamics of the social network in-degree effects → increasing, plateauing topological distance effects → strong trend to repetition and local 1 10 interaction f ( k ) ˆ semantic distance 0 10 → strong trend to homophily 0 50 100 150 200 k primarily “social”? social distance and degree
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 1 10 → strong trend to repetition and local interaction f ( d ) 0 ˆ 10 semantic distance −1 → strong trend to homophily 10 primarily “social”? −2 10 1 2 3 4 >4 d social distance and degree
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics Dynamics of the social network in-degree effects 2 → increasing, plateauing 10 topological distance effects 1 10 → strong trend to repetition and local interaction g ( δ ) ˆ semantic distance 0 10 → strong trend to homophily primarily “social”? −1 10 [0;.1] ].1;.2] ].2;.3] ].3;.4] ].4;.5] ].5;.6] ].6;.7] ].7;.8] ].8;.9] ].9;1] δ social distance and degree
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 −1 10 propension p ( d, δ ) → strong trend to repetition and local interaction −2 10 −3 10 semantic distance → strong trend to homophily −4 10 1 2 [0;0.2[ [0.2;0.4[ [0.4;0.6[ [0.6;0.8[ [0.8;1] 3 >3 social distance d primarily “social”? semantic distance δ social distance and degree
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 10 propension p ( d, k ) → strong trend to repetition and local 1 10 interaction 0 10 semantic distance −1 10 → strong trend to homophily −2 10 1 2 100 3 50 >3 0 primarily “social”? social capital k social distance d social distance and degree
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics Information flows: measures on the post network a Dyadic measures: 1 1 3 2 raw, weighted network, aggregated on 4 months b 1 2 attentional matrix a ... c → and total attention α a = 5 / 6 detachment matrix 2 3 1 1 f 2 “edge range”: 1 quantifying shortcuts 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: 1/6 1/4 3/4 2/3 raw, weighted network, aggregated on 4 months b 1/3 2/6 attentional matrix a ... c → and total attention α a = 5 / 6 1/5 detachment matrix 2/2 3/6 1/5 f 2/3 “edge range”: 1/3 quantifying shortcuts 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 4/3 3/2 raw, weighted network, aggregated on 4 months b 3 3 attentional matrix a ... c → and total attention α a = 5 / 6 5 detachment matrix 1 2 5 f 3/2 “edge range”: 3 quantifying shortcuts 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: 4 4/3 3/2 raw, weighted network, b aggregated on 4 months 3 attentional matrix a ... c → and total attention 1 α a = 5 / 6 f detachment matrix 3/2 5 2 3 5 e “edge range”: quantifying shortcuts 5/3 d
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics Information cascade Diffusion subgraphs 4 10 links nodes number of diffusion subgraphs a 3 20/02 b 26/02 d 10 20/02 19/02 2 10 c 1 10 An example of diffusion subgraph, a common “resource” and a set of 0 10 0 1 2 10 10 10 size of diffusion subgraphs citation links between blogs ⇒ heterogeneous cascade sizes
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 10 Number of tranmissions We focus on the total number of “transmissions” generated by blogs 1 10 with a given total attention α 0 10 a bit more “global”... −1 10 −2 −1 0 1 2 10 10 10 10 10 Total Attention α second transmissions: we focus on Larger active readership = > larger “later transmissions”, i.e. after a first number of diffusion links, yet not linearly transmission event
Recommend
More recommend
