Abstract
Sentiment analysis of social media data has become an essential research topic due to the widespread use of platforms such as Twitter for expressing public opinion. However, the short, informal, and context-dependent nature of tweets poses significant challenges for accurate sentiment classification. In this study, a comprehensive sentiment analysis framework is proposed using a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architecture. The proposed model integrates the local feature extraction capability of Convolutional Neural Networks (CNN) with the sequential and contextual modeling strength of Long Short-Term Memory (LSTM) networks. The framework is evaluated on a large-scale and balanced Twitter dataset containing over 690,000 tweets labeled as positive, negative, and neutral. Multiple deep learning models, including CNN, LSTM, Bidirectional Long Short-Term Memory (Bi-LSTM), and the proposed CNN-LSTM hybrid model, are systematically compared using standard evaluation metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrate that the CNN-LSTM hybrid model achieves the best overall performance, with an accuracy of 97.73%, precision of 97.74%, recall of 96.73%, and an F1-score of 97.73%, outperforming all baseline models. Furthermore, model interpretability is enhanced using SHapley Additive exPlanations (SHAP), which reveal that sentiment predictions are driven by semantically meaningful and sentiment-relevant words. The experimental findings suggest that the proposed approach provides a strong balance between predictive performance and interpretability for large-scale Twitter sentiment analysis tasks.