Where to Add Actions in Human-in-the-Loop Reinforcement Learning