A Hybrid CNN-LSTM Model with SMOTE for Enhanced Sentiment Analysis of Hotel Reviews

The growing reliance on online reviews as a critical decision-making tool in the hospitality industry underscores the need for robust sentiment analysis methodologies. Understanding customer feedback is essential for hotels to enhance service quality and maintain a competitive edge in an increasingly digital marketplace. However, traditional sentiment analysis models often encounter difficulties processing unstructured textual data, particularly when faced with class imbalances where positive reviews dominate, overshadowing critical negative feedback. To address these challenges, this study investigates integrating a hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) model with the Synthetic Minority Over-sampling Technique (SMOTE) to improve sentiment classification accuracy. Utilizing a dataset of 665 reviews from THE 1O1 Bandung Dago Hotel, the model leverages CNN’s capability to capture local features and LSTM’s strength in handling sequential dependencies, resulting in a more nuanced analysis of customer sentiments. The application of SMOTE effectively balances the dataset, addressing the class imbalance issue, which often skews sentiment classification. This approach improves predictive accuracy and provides actionable insights to enhance customer satisfaction strategies. The study achieved an overall classification accuracy of 77%, with precision at 78%, recall at 77%, an F1 score of 77.5%, and an AUC score of 0.81, reflecting discriminatory solid capability. Future research could focus on model optimization, multilingual sentiment analysis, aspect-based sentiment insights, and real-time sentiment monitoring to further refine customer feedback analysis and support strategic decision-making in the hospitality sector.