Within large communities, individuals sparsely interact with each others but set a tight releationshipwith a limited number of subjects only.These aggregations depend on the nature of the relationship, being guided by geographic neighbourhood,task sharing, homophily, and other agglomerative processes.
In social network analysis this is translated into the definition of multiple layers, and the actors social behaviour results in the creation of clusters of densely connected actors, loosely connected with actors of other groups. A question of interest is to identify the groups and reveal their formation mechanisms.
The correspondence between groups of subjects and the inner connection density, suggests the idea of extending the density-based approach for clustering non-relational data to the network framework. The nonparametric formulation of this approach associates clusters with high-density regions of the sample space. While a probabilistic notion of density is undefined for networks, this lack allows us to consider ad-hoc measures depending on the kind of aggregation mechanism one is interested to uncover.The proposed method allows to deal with very general relational structures such as the so-called multiplex networks - networks spanned on the same actors interacting through different relationships - for which very few methods have been proposed.
