Social networking websites provide a huge source of social data. We have built a multiagent simulation framework to handle this data and to simulate social interactions stochastically. Progress of the simulation is visualized using self-organizing maps. Based on the framework, we propose a recommender system that combines the traditional collaborative filtering and content-based methods. To evaluate the framework and the recommender system we collect music listening data from the Last.fm service and analyze two different example runs of the simulation. The framework proves to be functional but we are lacking history data which could be used to calibrate the free parameters in the simulation and validate its predictions.

is submitted