Limits...
Generating attributed networks with communities.

Largeron C, Mougel PN, Rabbany R, Zaïane OR - PLoS ONE (2015)

Bottom Line: Evaluating algorithms or comparing algorithms for automatic discovery of communities requires networks with known structures.Synthetic generators of networks have been proposed for this task but most solely focus on connectivity and their properties and overlook attribute values and the network properties vis-à-vis these attributes.In this paper, we propose a new generator for attributed networks with community structure that dependably follows the properties of real world networks.

View Article: PubMed Central - PubMed

Affiliation: Hubert Curien Laboratory, University of Lyon, Saint-Étienne, France.

ABSTRACT
In many modern applications data is represented in the form of nodes and their relationships, forming an information network. When nodes are described with a set of attributes we have an attributed network. Nodes and their relationships tend to naturally form into communities or clusters, and discovering these communities is paramount to many applications. Evaluating algorithms or comparing algorithms for automatic discovery of communities requires networks with known structures. Synthetic generators of networks have been proposed for this task but most solely focus on connectivity and their properties and overlook attribute values and the network properties vis-à-vis these attributes. In this paper, we propose a new generator for attributed networks with community structure that dependably follows the properties of real world networks.

No MeSH data available.


Evolution of structural measures for  varying. is fixed to 20 to ensure that the number of within edges remains higher than the number of between edges. The evolution of the average clustering coefficient is presented on the left side. The evolution of the modularity is presented on the right side.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4404059&req=5

pone.0122777.g005: Evolution of structural measures for varying. is fixed to 20 to ensure that the number of within edges remains higher than the number of between edges. The evolution of the average clustering coefficient is presented on the left side. The evolution of the modularity is presented on the right side.

Mentions: In the aim to degrade the community structure in such a way to have a number of within edges lower than the number of between edges, we performed experiments in varying the values of the parameter which controls the maximum number of edges connecting a new vertex to vertices out of its community. Table 6 presents several structural measures for and . As shown in Fig 4 presenting the corresponding graphs, the number of edges between the communities increases. Fig 5 presents the average clustering coefficient (left side) and the modularity (right side) for ranging between 0 and 20 and to respect . As expected, these results indicate that the community structure is degraded when parameter increases. One can also notice that even if the community structure is degraded, the obtained average clustering coefficient remains higher than the one obtained in an Erds-Renyi random graph having the same number of vertices and edges (i.e., the random clustering coefficient measure).


Generating attributed networks with communities.

Largeron C, Mougel PN, Rabbany R, Zaïane OR - PLoS ONE (2015)

Evolution of structural measures for  varying. is fixed to 20 to ensure that the number of within edges remains higher than the number of between edges. The evolution of the average clustering coefficient is presented on the left side. The evolution of the modularity is presented on the right side.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4404059&req=5

pone.0122777.g005: Evolution of structural measures for varying. is fixed to 20 to ensure that the number of within edges remains higher than the number of between edges. The evolution of the average clustering coefficient is presented on the left side. The evolution of the modularity is presented on the right side.
Mentions: In the aim to degrade the community structure in such a way to have a number of within edges lower than the number of between edges, we performed experiments in varying the values of the parameter which controls the maximum number of edges connecting a new vertex to vertices out of its community. Table 6 presents several structural measures for and . As shown in Fig 4 presenting the corresponding graphs, the number of edges between the communities increases. Fig 5 presents the average clustering coefficient (left side) and the modularity (right side) for ranging between 0 and 20 and to respect . As expected, these results indicate that the community structure is degraded when parameter increases. One can also notice that even if the community structure is degraded, the obtained average clustering coefficient remains higher than the one obtained in an Erds-Renyi random graph having the same number of vertices and edges (i.e., the random clustering coefficient measure).

Bottom Line: Evaluating algorithms or comparing algorithms for automatic discovery of communities requires networks with known structures.Synthetic generators of networks have been proposed for this task but most solely focus on connectivity and their properties and overlook attribute values and the network properties vis-à-vis these attributes.In this paper, we propose a new generator for attributed networks with community structure that dependably follows the properties of real world networks.

View Article: PubMed Central - PubMed

Affiliation: Hubert Curien Laboratory, University of Lyon, Saint-Étienne, France.

ABSTRACT
In many modern applications data is represented in the form of nodes and their relationships, forming an information network. When nodes are described with a set of attributes we have an attributed network. Nodes and their relationships tend to naturally form into communities or clusters, and discovering these communities is paramount to many applications. Evaluating algorithms or comparing algorithms for automatic discovery of communities requires networks with known structures. Synthetic generators of networks have been proposed for this task but most solely focus on connectivity and their properties and overlook attribute values and the network properties vis-à-vis these attributes. In this paper, we propose a new generator for attributed networks with community structure that dependably follows the properties of real world networks.

No MeSH data available.