Automatic resolution of model merging conflicts using quality-based reinforcement learning