EN FR
EN FR
DANTE - 2012




Bibliography




Bibliography


Section: New Results

Community detection: dynamic, overlapping, fuzzy

Community, a notion transversal to all areas of Social Network Analysis, has drawn tremendous amount of attention across the sciences in the past decades. Numerous attempts to characterize both the sociological embodiment of the concept as well as its observable structural manifestation in the social network have to this date only converged in spirit. No formal consensus has been reached on the quantiffable aspects of community, despite it being deeply linked to topological and dynamic aspects of the underlying social network.

The DANTE team proceeded results on several aspects of community detection is large scale networks.

  • Presenting a fresh approach to the evaluation of communities, we introduces and builds upon the cohesion [8] , a novel metric which captures the intrinsic quality, as a community, of a set of nodes in a network. The cohesion, defined in terms of social triads, was found to be highly correlated to the subjective perception of communitiness through the use of a large-scale online experiment in which users were able to compute and rate the quality of their social groups on Facebook. The use of the cohesion proves invaluable in that it offers non- trivial insights on the network structure and its relation to the associated semantic. The use of the cohesion was use for example in order to study Agreement Groups in the United States Senate [35] .

  • Overlapping community detection is a popular topic in complex networks. As compared to disjoint community structure, overlapping community structure is more suitable to describe networks at a macroscopic level. Overlaps shared by communities play an important role in combining different communities. In this paper, two methods are proposed to detect overlapping community structure. One is called clique optimization, and the other is named fuzzy detection. Clique optimization aims at detecting granular overlaps. The clique optimization method is a fine grain scale approach. Each granular overlap is a node connected to distinct communities and it is highly connected to each community. Fuzzy detection is at a coarser grain scale and aims at identifying modular overlaps. Modular overlaps represent groups of nodes that have high community membership degrees with several communities. A modular overlap is itself a possible cluster/sub-community [7] , [38] .