ARCHIVES

In the era of rapid digital transformation, user-generated content has emerged as a powerful factor influencing consumer decision-making, especially within the hospitality industry. Online hotel reviews, shared across various booking platforms and social media channels, provide rich and valuable insights into customer experiences, satisfaction levels, and service quality. However, the exponential growth in the volume of such unstructured textual data makes manual analysis not only time-consuming but also inefficient and impractical for large-scale applications. To address this challenge, this research proposes an automated sentiment analysis framework based on advanced deep learning techniques to effectively classify hotel reviews into positive and negative sentiment categories. The proposed system leverages Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM) models, which are well-suited for handling sequential textual data and capturing contextual dependencies within sentences. These architectures enable the model to understand complex linguistic patterns, including long-term dependencies and semantic relationships in natural language. A comprehensive data preprocessing pipeline is implemented to enhance model performance. This includes text cleaning (removal of noise such as punctuation, special characters, and stopwords), tokenization, and sequence padding, which transforms raw textual data into a structured numerical format suitable for neural network training. Additionally, word embedding techniques are employed to represent words in dense vector form, improving the semantic understanding of the models. The models are trained and evaluated using a large-scale dataset of hotel reviews, ensuring robustness and generalizability of the proposed approach. Experimental results indicate that the LSTM model achieves superior performance with an accuracy of approximately 90.7%, slightly outperforming the BiLSTM model, which also demonstrates competitive results. The evaluation metrics, including accuracy and loss analysis, confirm the effectiveness of the proposed models in sentiment classification tasks