In this paper, we consider a joint design of the user clustering and content caching in the cache-enabled heterogenous network (HetNet) in which users in the network have distinct content preferences. The joint clustering and caching in the HetNet relies on multitude of factors, such as channel gains in all links, which may not be fully known in practice. Besides, clustering and caching may exhibit a fundamental tradeoff between the content hit probability and the spectral efficiency. We are therefore motivated to tackle this challenging problem by the deep reinforcement learning (DRL). In particular, the deep deterministic policy gradient (DDPG) algorithm is employed to manage the dynamics of clustering and caching in the HetNet with a sizable action space. Simulation results are presented to demonstrate the effectiveness of the proposed algorithm.