Génération multi-agents de réseaux sociaux
Audren Bouadjio-Boulic · HAL (Le Centre pour la Communication Scientifique Directe) · 2021
In this thesis, we study the reproduction of social networks using an Agent-Based simulation. Two distinct approaches are often used in order to generate social networks. The first consists in mathematically replicating the topological properties of the network. The other is to recreate the social behaviors leading to network creation. The advantage of the latter is that it is understandable and tractable by users from any domain. The drawback of this approach is that the networks generated may not have realistic topologies. On the other hand, the first approach leads to good topological results. Unfortunately, the easy-to use models leads to quantitative results whereas the qualitative results require the use of harder models. Our model stands between these two approaches. The generation of the network is the result of an agent-based simulation where agents follow social-interpretable rules. The validation of the result is performed on the generated network topology. This method is related to complex systems: microscopic interaction phenomena lead macroscopic results which are generated networks. First, a general framework is theoretically defined, offering concepts and common mechanisms shared by model instances. Experiments are performed to analyze agent behavior and the resulting network topologies. Two instances of the model are then created and finally the parameter space of the model is explored using a genetic algorithm. Each of these models offer advantages and drawbacks concerning network generation. We compare the results of those models with different social networks. In this work, we offer an original way to generate networks which is both easy to use and precise in the topologies generated.