This study introduces a strong hybrid movie recommendation system that was developed using a two-stage methodology: a deep learning based semantic-genre recommendation engine and a machine learning based sentiment classification module. Using the Term Frequency-Inverse Document Frequency (TF-IDF) method, user reviews from the IMDb dataset are first converted into vectors. The Light Gradient Boosting Machine (LightGBM) algorithm is then used to perform sentiment classification. Class imbalance is addressed by the Synthetic Minority Over-sampling Technique (SMOTE), which gives the model a maximum classification accuracy of 83.32%. Sentence-BERT embeddings and FAISS (Facebook AI Similarity Search) indexing are used in the second section to enable real-time semantic similarity searches over a large collection of movie metadata. Contextual relevance and emotional engagement are increased when sentiment and genre preferences are incorporated into the recommendation pipeline. The final model provides a high-accuracy, scalable way to provide highly customized movie recommendations.