Identification de r\^oles communautaires dans des r\'eseaux orient\'es appliqu\'ee \`a Twitter

TL;DR

This paper extends community role identification methods to directed networks and applies it to Twitter, revealing specific roles of users like social capitalists through an unsupervised approach.

Contribution

It generalizes existing community role measures to directed networks and introduces an unsupervised method for role determination, demonstrated on Twitter data.

Findings

Identified specific roles of social capitalists in Twitter.

Extended community role measures to directed networks.

Demonstrated effectiveness of the method on real social media data.

Abstract

The notion of community structure is particularly useful when analyzing complex networks, because it provides an intermediate level, compared to the more classic global (whole network) and local (node neighborhood) approaches. The concept of community role of a node was derived from this base, in order to describe the position of a node in a network depending on its connectivity at the community level. However, the existing approaches are restricted to undirected networks, use topological measures which do not consider all aspects of community-related connectivity, and their role identification methods are not generalizable to all networks. We tackle these limitations by generalizing and extending the measures, and using an unsupervised approach to determine the roles. We then illustrate the applicability of our method by analyzing a Twitter network.We show how our modifications allow discovering the fact some particular users called social capitalists occupy very specific roles in this system. --- La notion de structure de communaut\'es est particuli\`erement utile pour \'etudier les r\'eseaux complexes, car elle am\`ene un niveau d'analyse interm\'ediaire, par opposition aux plus classiques niveaux local (voisinage des noeuds) et global (r\'eseau entier). Le concept de r\^ole communautaire permet de d\'ecrire le positionnement d'un noeud en fonction de sa connectivit\'e communautaire. Cependant, les approches existantes sont restreintes aux r\'eseaux non-orient\'es, utilisent des mesures topologiques ne consid\'erant pas tous les aspects de la connectivit\'e communautaire, et des m\'ethodes d'identification des r\^oles non-g\'en\'eralisables \`a tous les r\'eseaux. Nous proposons de r\'esoudre ces probl\`emes en g\'en\'eralisant les mesures existantes, et en utilisant une m\'ethode non-supervis\'ee pour d\'eterminer les r\^oles. Nous illustrons l'int\'er\^et de notre m\'ethode en l'appliquant au r\'eseau de Twitter. Nous montrons que nos modifications mettent en \'evidence les r\^oles sp\'ecifiques d'utilisateurs particuliers du r\'eseau, nomm\'es capitalistes sociaux.