Deep Learning for Sentiment Analysis: Combining CNN and LSTM for Enhanced Performance on Facebook Data

This research evaluates the effectiveness of different sentiment analysis techniques, especially the hybrid CNN-LSTM model, on a large dataset of 100,000 Facebook posts and comments. For accuracy, precision, recall, and F1-score evaluation, many metrics are in use for comparison with the traditional machine learning models (Naive Bayes, SVM) and other deep learning models such as the standalone CNN and LSTM vis-à-vis the CNN-LSTM model. The CNN-LSTM model showed the highest performances across all metrics: accuracy of 0.91, precision of 0.90, recall of 0.90, and an F1 score of 0.90. Detailed error analysis has shown the problems in sarcasm detection and, to a small extent, the detection of subtly formulated sentiment, while confusion matrix analysis also shows that the neutral class has systematically been mislabeled as positive or negative. These findings point out possible areas one needs to look at for improvements by the inclusion of attention mechanisms. The results of this study emphasize the CNN-LSTM model’s practical applicability to customer feedback analysis, brand monitoring, and market research. Advanced techniques and diverse datasets can be used in further research to improve the robustness of the model and its generalizability.