Title: SentimentLens: Bridging the Gap Between Sentiment and Ratings Through Dual-Modality in Hospitality

Abstract:

Online travel agencies produce massive quantities of user-generated hotel reviews, presenting significant potential for gaining a comprehensive understanding of traveler experiences. Nevertheless, converting this unstructured textual data into structured, actionable insights remains a complex challenge. This study introduces SentimentLens, a scalable analytical framework leveraging Aspect-Based Sentiment Analysis to extract knowledge from unstructured hotel feedback and organize it into interpretable service categories. By combining aspect term extraction, sentiment classification, semantic category assignment, and multi-level analytical modules, the system facilitates evaluations at the regional, hotel, and service category levels. It is engineered to function effectively across diverse geographic locations and hospitality environments.

To validate its practical application, SentimentLens was applied to a substantial real-world dataset comprising more than 10,000 publicly accessible hotel reviews. Comprehensive analysis demonstrates how traveler sentiment fluctuates across different regions, service types, and hotel profiles. Furthermore, the study employs importance–performance and entropy-based analyses to reconcile textual sentiment with numerical ratings. This cross-modal approach uncovers latent operational conflicts, structural inconsistencies in service quality, and high-impact opportunities for improvement. The findings indicate that SentimentLens successfully converts large-scale unstructured reviews into actionable intelligence, thereby aiding data-driven decision-making for tourism policy and hospitality management. Although illustrated through a national case study, the proposed framework is adaptable to other destinations and service sectors reliant on user reviews.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC