Edge caching technology is crucial for improving user experience and reducing the load on origin servers by deploying multiple caching nodes at network edges. Given the limited capacity of these edge caches, managing the space effectively to accommodate dynamically changing user requests is paramount. Traditional optimization and heuristic algorithms, while sometimes offering viable solutions, generally lack adaptability to dynamic environments, leading to potentially suboptimal performance. To this end, we propose a hierarchical clustering tree reinforcement learning algorithm for efficient content caching. Specifically, to overcome the inefficiencies of traditional reinforcement learning in large action spaces, the long short-term memory network is employed to preprocess request content sets, organizing them into a balanced hierarchical clustering tree. Then, the reward function is designed to maximize the average hit rate of requested video content. Furthermore, we utilize a simple recurrent unit to build a state tracker that continuously gathers and records interaction trajectories. Experimental results on the real iQIYI dataset confirm the effectiveness of our proposed approach.