Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Poster presented at EGC 2011
1. Point of View Based Clustering of
Socio-Semantic Networks
Influencing the communities dectection process in socio–
semantic networks using points of view
Authors Socio–semantic networks: enhancing the structure Points of View
Juan David CRUZ GOMEZ with semantics Given a graph G (V, E), let FV be
Cécile BOTHOREL Social Graph Information the set of semantic features of the
Node degree, node centrality,
François POULET node betwenness, prestige actors of the network, and let
Walks and paths, relationships ∗
strenght, types of relationship
This is t he st ruct ural
informat ion t he
Inf o rmat io n
abo ut t he act o r FV ∈ P (FV ) FV , be a non–empty
net work Na me
Density of the graph, geodesics,
Partenaires distance and diameter,
Type
Da te of inclusion
set of features to be used to
connectivity of the graph Socio-Semantic into the network
define the point of view P oV .
Semantic Information Network
Role of the actors, actor's name/
A point of view is defined as the
filliation, actor's position
Type of relationship, relationship
set of all instances derived from
Point s of view are
statistics (date, evolution)
Evolution of the network,
creat ed from t hese
feat ures
the set FV :
contexts of the network
|V |
updates, other features of the
network P oVFV∗ = ξi
i=1
By using the structural information and the semantical information in a conjoint where ξi is the binary vector
way it is possible to extract non–evident information and use it to analyze the (instance) assigned to the node i.
network from different perspectives.
Model Inputs Phase 1: semantic clustering
Point of View
Feature 1 Feature 2 Feature 3
The point of view 1 2
is a set of binary 1 2
Node 1 1 0 0
vectors representing
Node 2 0 1 0
a subset of features 25 5 4 3
Node 3 0 1 1
from the socio- 6
Node 4
.
1
.
1
.
0
. semantic network and 27
7
. . . .
assigned to each actor
. . . .
Node 29 0 1 1
in it. 9
26 8
Semantic
Social Graph
Clustering 10 11 28
12 13
1 2
25 5 4 3
6
16 15
27 The social graph is the 14
7
representation of the 17 18
26 9 8
relationships between
19 20
10 11 28
12 13
the actors in the socio- 29 22
24 21
semantic network. Using Self-Organizing Maps [1] the nodes 23
16 15 14
17 18 19 20 are clustered from a semantic perspective Each node belongs to one semantic group
29 22
24 21
23
Phase 2: communities detection
Algorithm General Steps
3 1
4 2
The nodes semantically clustered
1 25 5
according to the point of view 4 3
1 2 6
Each node in the network is 25
1
20
1
27
2 1
20 5 4 20 3
assigned to one semantic group 6 1
20 1 20 7
1 27 1
The weights of the edges of 1
1 7 26 9 8
3 the network are changed according 26 9 1
8
to the semantic groups 1 1 20
20
10 11 12 13
10 1 11 28
12 1 13 28
The communities are found using 20
1
1
1
1 1
20
the Fast Unfolding algorithm [2] on 20
16 1 20 1 15 1 16 15
4 the social graph augmented with 17
14 14
18 19 1
20 17 18
semantic weights 29
1
1 1
20
1
22 20 1
19 20
24
23 21 29 22
24 21
The weights are changed according 23
References the semantic distance The final communitites are structurally
and semantically similar
[1] T. Kohonen, Self-Organizing Maps. Springer,
1997. Conclusion
[2] V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and
E. Lefebvre, “Fast unfolding of communities in large Assigning weights derived from the results of the semantic clustering to the edges, the semantic
networks,” Journal of Statistical Mechanics: Theory information is included into the community detection process and the two types of data are
and Experiment, vol. 2008, no. 10, p. P10008, 2008 merged to find and visualize a social network from a selected point of view.
Contact : juan.cruzgomez@telecom-bretagne.eu, http://www.telecom-bretagne.eu/