Résumé: Social media needs social networks to disseminate information among the people who like to interact with each other. Identification of the most prominent influencers is a crucial problem for influence diffusion in applications like viral marketing. Most of the previous works on influence diffusion studied the topological nature of the users but ignored the effects of communities within a network. This paper proposes a new method called the TSGC method, where the whole graph is divided into different non-overlapping communities. Each of the communities is taken as a subgraph by ignoring the connecting links between them. The most influential users are identified by using each node's local and global centrality measures in the given subgraph of a graph. Finally, the ranking of each node is performed by calculating the of each node within a whole network. Experimental results on six datasets confirm that the proposed TSGC method outperforms many well-known existing methods in terms of influence diffusion phenomenon under both the LT and IC models. This paper also proposes a new model where an active promoter may lose his influencing potential over time, go to a recovered state where he is no longer active or can activate others, and then go to a susceptible state where he is prone to getting influenced in the future. A user who is influenced by an active user can also become an active user. We termed this model the ICIRS model. This model undergoes influence diffusion in continuous time, unlike discrete-time steps as focused in most of the existing papers. Our experimental evaluations on the datasets reveal that the ICIRS model performs non-progressive influence diffusion.

Langue: Anglais