In the era of increasingly abundant digital content, personalized recommendation systems play a critical role in helping users efficiently identify relevant movies. However, traditional approaches such as Collaborative Filtering (CF) and Content-Based Filtering (CBF) continue to suffer from data sparsity, cold-start limitations, and unstable clustering performance. To address these constraints, this study proposes a novel hybrid recommendation framework that integrates Harmony Search (HS) optimization with Fuzzy C-Means (FCM) clustering inside a Hybrid Filtering (HF) architecture. Using a subset of the MovieLens dataset consisting of 560 users who rated the same 37 movies, 60% of the rating values were randomly removed to simulate sparse conditions. HS is employed to optimize the initialization of FCM centroids, improving clustering stability and reducing susceptibility to local minima. The resulting clusters are then leveraged in a hybrid combination of CF and CBF to generate final predictions. Experimental results indicate that the optimal configuration (num_cluster = 4, m = 1.5, α = 0.7) achieves RMSE = 0.8974, MAE = 0.7011, Precision = 0.7515, and Recall = 0.4628. Compared to baseline models, the proposed HS–FCM–HF framework improves RMSE by 37.3% over CBF-only and maintains 7.4% better Precision than CF-only, demonstrating stronger robustness and balanced performance under high sparsity. These findings highlight the theoretical and practical value of integrating metaheuristic optimization with hybrid filtering to enhance both accuracy and generalization. Future work may incorporate multimodal features or real-time adaptive mechanisms to further strengthen personalization capability.