In the last decade, researchers has focused their studies on the mathematical relation between the Quality of Service (QoS) and the user Quality of Experience (QoE). This paper investigates the problem of modelling the user QoE feedback in the next generation networks. The problem has been formulated and solved using a reinforcement learning technique. The proposed approach is innovative since it does not require an explicit knowledge of the mathematical model describing the network dynamics or the QoS/QoE relationship since it is learnt on-line. Simulation results shows that the proposed solution can adapt dynamically to the user behavior.