The rapid increase in available movie content makes it difficult for users to choose films that align with their preferences. To address this issue, this study proposes a hybrid movie recommendation system that employs content-based filtering (CBF) and collaborative filtering (CF) together, using a Weighted Hybrid Filtering (WHF) method. The collaborative part of the recommendation system utilizes a Restricted Boltzmann Machine (RBM) to model latent trends within sparse user-item interactions and the content-based portion of the recommendation system uses TF-IDF to model textual features of movies, and cosine similarity to measure relevance in similarity. Hybrid recommendation scores are generated using various weight ratios, with the optimal performance achieved at a 0.2 weight for CBF and 0.8 for RBM-based CF. To enhance interpretability and refine recommendations, the hybrid scores are further classified into binary decisions indicating whether a movie is recommended (1) or not recommended (0) using a Recurrent Neural Network (RNN). The RNN is adopted to leverage its ability to model sequential dependencies and retain contextual information across user preferences. The dataset used consists of 854 Netflix and Disney+ movies and 34,086 user-generated reviews collected via web crawling. The experiments indicate that the WHF model has a Mean Absolute Error (MAE) of 0.0204 and a Root Mean Square Error (RMSE) of 0.0363. The highest classification accuracy of 85.96% was achieved with the Adamax optimizer at the learning rate of 0.00223. These results conclude combining hybrid filtering with neural-based classification effectively improves both the accuracy and relevance of movie recommendations.