Hybrid CNN-LSTM approach for sentiment analysis on IMDB movie reviews
Rahul Kumar Gupta, Binayak Ojha, Surya Prasad Yadav, Abhinav Kumar Singh, Prasant Kumar Dash, Aadarsh Kumar Singh · 2025
Paper advances a hybrid model developed through an approach based on convolutional neural networks and LSTM for sentiment analysis tasks that would focus on the IMDB movie review data. Aim of this paper will be to distinguish the opinion into positive and negative emotions based on the strengths of both CNNs for the feature extraction and LSTMs for sequence modeling. We have conducted a vast number of experiments compared CNN-LSTM hybrids with basic models (LSTM model, CNN model, and Logistic Regression with term frequency-Inverse document frequency). Additionally, we carried out an ablation study to note the importance of various modules while building up the model, and done Hyperparameter optimization for the optimization of model accuracy, and Conducting adversarial attacks to test robustness on the models. The LIME and SHAP frameworks improved model interpretability, adding much-needed transparency to the classifications. The Proposed model CNN-LSTM has enhanced accuracy, recall, F1-score and precision comparison to baselines.