In the realm of recommender systems, enhancing the quality of recommendation lists has become a focal point for researchers. This paper presents a novel approach integrating clustering structures with Graph Convolutional Network (GCN) techniques to improve recommendation quality. Initially, we employ a hierarchical tree structure to cluster similar users and items based on energy-based similarity measures. This allows for a more accurate modeling of user and product groups. We then construct graphs representing user relationships (SU-Graph) and item relationships (SI-Graph) based on these clusters, as well as a graph derived from the user-item rating matrix. Utilizing this framework, we train a GCN to predict user ratings for previously unseen items, significantly enhancing the accuracy of recommendations. Finally, we refine the recommendation lists by balancing precision and diversity, ensuring users receive suggestions that are both relevant and varied. Experimental results on the MovieLens dataset validate the effectiveness of our proposed approach, demonstrating substantial improvements over traditional methods.