Enhancing Sentiment Analysis via Advanced Deep Learning Techniques

Sentiment analysis has become a fundamental task in natural language processing, facilitating the automatic extraction of subjective information from text. This study evaluates the effectiveness of various sentiment analysis models by comparing widely adopted machine learning and deep learning approaches. Specifically, we implement and assess Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks independently before integrating them into a hybrid CNN-LSTM architecture. Additionally, we benchmark these deep learning models against traditional machine learning methods, including Naïve Bayes, Decision Tree, K-Nearest Neighbors (KNN), XGBoost, and AdaBoost, to identify the most effective sentiment classification technique. Experiments are conducted using the IMDb movie review dataset, where models are evaluated based on recall, precision, F1-score, and accuracy. The results demonstrate that deep learning models substantially outperform traditional machine learning approaches. Among all models, the CNN-based architecture achieves the highest F1-score (0.8873) and accuracy (0.8861), highlighting its superior capability in capturing sentiment-related features. The CNN-LSTM hybrid and LSTM-based models also yield strong performance, emphasizing the importance of sequential dependencies in sentiment classification. Conversely, traditional machine learning models exhibit lower effectiveness, with Naïve Bayes emerging as the most competitive among them but still trailing behind deep learning approaches. These findings confirm the robustness of deep learning models for sentiment analysis, with CNN-based architectures proving to be both highly accurate and computationally efficient.