Deep Opinion Representation Learning for Dynamic Hospitality Demand Estimation from Customer Review Ecosystems

Abstract

Sentiment analysis has emerged as a crucial application in Natural Language Processing (NLP), enabling the identification and interpretation of opinions and emotions expressed in textual data. With the rapid expansion of social media and online review platforms, customers actively share their experiences regarding products and services, generating large volumes of opinion-rich data. Analyzing this information has become essential for businesses to understand customer preferences, enhance service quality, and remain competitive in a technology-driven environment. Various algorithms have been explored for sentiment classification, including both traditional Machine Learning (ML) techniques and advanced Deep Learning (DL) models. In this work, DL approaches such as Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs) were applied to hotel review data, achieving accuracy rates of 86% and 84%, respectively, demonstrating their ability to capture contextual and sequential patterns effectively. Additionally, traditional classifiers such as Naïve Bayes (NB), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM) were evaluated, producing accuracies of 75%, 71%, 82%, and 71%. The comparative results indicated that DL-based models outperformed conventional methods in sentiment prediction tasks, providing more accurate and reliable insights from customer reviews. These findings highlight the effectiveness of DL techniques in improving business intelligence by enabling better understanding of customer feedback and supporting informed decision-making