A Hybrid Deep Learning Framework for Hotel Rating Systems: Integrating Word2Vec, TF-IDF, and Bi-LSTM With Attention Mechanism

The hospitality industry faces a persistent challenge in accurately gauging customer sentiment from online reviews, often resulting in a disparity between ratings and actual experiences. Conventional hotel rating systems struggle to capture the nuanced opinions expressed in user-generated content. To address this issue, our research presents a novel hotel rating model leveraging natural language processing (NLP) techniques. This model directly analyzes and synthesizes customer feedback from online reviews to generate dynamic and precise hotel ratings. By incorporating Word2Vec, term frequency-inverse document frequency (TF-IDF), and a three-layer bidirectional long short-term memory (Bi-LSTM) network with a lightweight self-attention mechanism, our approach adeptly identifies sentiment tendencies within diverse review data. The proposed hotel rating model outperformed baseline models across experiments, demonstrating superior adaptability, scalability, and robustness. Notably, it achieved an accuracy of 0.9123 and an area under the curve (AUC) of 0.9536 on diverse datasets, and its performance consistently improved with increased data volumes, reaching 0.9376 accuracy and 0.9612 AUC at 75% data volume. Its low variability in performance metrics across multiple runs highlights its reliability for sentiment analysis in the hospitality sector.