Hotel Reviews Classification and Review-based Recommendation Model Construction using BERT and RoBERTa

Abstract

Personalization plays a crucial role in significantly enhancing customer satisfaction within the hotel industry. Customers, with their unique preferences, often rely on previous customer reviews when selecting a suitable hotel. Therefore, personalization can simplify this process by providing curated lists of hotel recommendations. Our research analyzed six hotel aspects: Value, Accessibility, Service, Room, Cleanliness, and Sleep Quality. We utilize transformer-based NLP models, namely BERT and RoBERTa, to accomplish the analysis. We propose a review classification method to identify customer preferences for each aspect and create a review-based recommendation system for hotel suggestions. To assess performance, we utilize three randomly selected reviews as inputs for both the classification model and the recommendation system. Our findings demonstrate that BERT outperformed RoBERTa in review classification, achieving an accuracy score of 0.8963 and a macro F1 score of 0.83. On the other hand, when constructing a review-based recommendation, RoBERTa proved superior to BERT, with the highest cosine similarity score of 0.99917. Based on our research, we recommend that the hotel industry consider leveraging NLP models, such as BERT and RoBERTa, to create effective personalization strategies. Our research contributes valuable scientific insights into the application of NLP models for creating personalized experiences within the hotel industry.

Topics & Concepts