In hot strip rolling mills, the looper control system is automated. However, the looper's behavior tends to be unstable in threading. Therefore, human expert always intervenes and stabilizes the looper's behavior by tuning PID gains and interposing manipulation variable of looper control system. In this paper, we propose a method based on the recurrent neural network to express PID gains tuning action by human. Furthermore, we propose two methods to update the model by learning. To check the effectiveness of the proposed learning methods, numerical simulation applied to the looper height control is carried out.