Etude de la Maximisation de l'Influence dans les Réseaux Sociaux

Cédric Lagnier (1) , Gaussier Eric (1)

1 AMA - Analyse de données, Modélisation et Apprentissage automatique [Grenoble]

Influence maximization is a NP-hard problem depending on the diffusion of information in social networks. The Greedy hill climbing algorithm have been proved a good approximation if the influence fonction we try to optimize is submodular, which is the case for standard diffusion models.We present a diffusion model not equivalent to standard models for which the influence function is not submodular. Then we propose, using toy graphs and a real social network, a study of different influence maximization algorithms on this model and on the standard model IC: some basic heuristics, the greedy hill climbing method, a generalization of the greedy method and an optimization method for submodular functions. We show that even if the influence function is not submodular, the greedy algorithm obtain good results while being able to scale efficiently.

Mots clés: Social Networks; Information Diffusion; Machine Learning

Domaines: Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]

Cédric Lagnier, Gaussier Eric. Etude de la Maximisation de l'Influence dans les Réseaux Sociaux. 4ième conférence sur les modèles et l'analyse des réseaux : Approches Mathématiques et informatiques, Oct 2013, Saint-Etienne, France. pp.32. ⟨hal-00881536⟩