This paper presents a hybrid deep learning framework for movie recommendation that leverages real-time Twitter data to address the limitations of static collaborative filtering. We propose a deep autoencoder architecture augmented with social context features (e.g., sentiment, trends) to model dynamic user preferences. Evaluated on the MovieTweetings dataset (200K ratings), our system reduces RMSE by 8.1% over SVD and 12.3% over k-NN, while outperforming recent GNN and transformer baselines. The study advances recommender systems by demonstrating the viability of social media integration, with implications for real-time personalization.