Hierarchical Reinforcement Learning for Modeling User Novelty-Seeking Intent in Recommender Systems