Advanced Agent-Based Modeling for Social Networks

Agent-based modeling and simulation have been successfully applied to problems emerging from social sciences and could be profitably used also for the online social networks. However, the tools presently available for agent-based modeling do not offer specific support for social network models. In the present work, we present a unified conceptual framework to develop both novel agent-based and traditional social network models. This conceptual framework is essentially a meta-model to express the other models. In addition, we designed a domain-specific language to formulate the models in an executable way, so that simulations can be performed effortlessly. The language aims at being expressive and powerful for those with a strong background in computing, and yet simple and easy to learn for those with different expertises. We also developed a software platform that can execute such models in an agent-oriented context, providing effective support for large networks. Moreover, the platform hides most of the complexity of running the simulations on remote server-class machines. We validated out approach by translating several traditional models in our meta-model, verifying that the expected features of the models are maintained. The results show that our approach is successful in providing a friendly and easy environment to perform agent-based simulations over social networks, simulations that are of interest both to develop models and to study the results of the models themselves. Then, considering the favorable results we obtained, we applied our platform to the still open problem of creating an entirely distributed social networking system, which, as compared to the centralized ones, yields relevant advantages as far as privacy and resilience are concerned. We developed several models to help us in the understanding of the many issues that a P2P social networking system would have when deployed, and specifically of the well-known issue of the availability of rare resources. Through simulations, we found some criteria for the design of distributed social networks and some operation conditions which may result in a satisfactory user experience in terms of reduced delays in the propagation of information. Consequently, these results allow us to develop now a distributed social networking system optimized by means of our simulations.