Understanding audience sentiment toward films is essential for filmmakers, marketers, and content creators. This paper presents a deep sentiment analysis approach that enhances movie review classification through an ensemble of advanced LSTM-based models. Specifically, the proposed architecture combines Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Convolutional Neural Networks (CNN) to address the contextual and sequential complexity of movie review texts. Traditional sentiment analysis techniques often fall short in capturing nuanced expressions and emotional subtleties. In contrast, our ensemble method, using weighted soft voting, leverages the strengths of each model to deliver more accurate and generalized sentiment predictions. Trained and evaluated on the IMDb dataset, the ensemble achieves an accuracy of 87%, outperforming individual models. This performance demonstrates the effectiveness of integrating multiple deep learning architectures for robust sentiment classification. The results offer valuable insights for film industry professionals seeking to understand and act on public opinion more effectively.