Exploring CNN, LSTM and CNN-LSTM for Sentiment Analysis in Film Reviews

With the widespread adoption of online platforms and user-generated content, the sentiment analysis of movie reviews has become a valuable task for academic research and industry applications. This paper systematically studies three main deep learning methods: convolutional neural network (CNN), long-term and short-term memory (LSTM) and its hybrid model CNN-LSTM. CNN captures local text features, while LSTM models long-term dependencies in the sequence. Hybrid models combine their advantages to provide higher accuracy and context understanding. Through the comparative analysis of data sets such as IMDb, Amazon and Twitter, the CNN-LSTM model has always shown excellent performance, with an accuracy rate of up to 98.09%. However, there are still challenges in dealing with irony and domain generalization. Future improvements may focus on integrating attention mechanisms and explainable artificial intelligence technology. This review provides valuable insights for the development of a robust, explainable and scalable sentiment analysis system for film reviews, and highlights potential directions for further enhancing model adaptability in real-word application.