Understanding how service attributes are associated with customer satisfaction is vital in the hospitality industry, where customer experiences differ by brand tier, location, and property characteristics. This study analyzes 110,955 Yelp reviews from 2,952 hotels to examine the relationships among review- derived service attributes, attribute-level sentiment, attribute-level sentence frequency, and satisfaction ratings. In the text-processing stage, reviews are segmented into sentence-level units using sentence- boundary rules and tokenization. Service attributes are extracted using LDA topic modeling, and six categories are identified: Hotel Facilities, Room Comfort, Customer Service, Breakfast Service, Entertainment & Family, and Dining Experience. VADER sentiment analysis is then applied at the sentence level to measure the evaluative tone associated with each attribute. These measures are aggregated into review-attribute-level frequency and sentiment variables. A multilevel ordered logit model examines how these variables are associated with ordinal satisfaction ratings while accounting for the nesting of reviews within hotels. The results show that Customer Service and Room Comfort have strong positive average associations with satisfaction, although their effects vary across hotel contexts. Additional robustness checks that account for review length and review-yea r effects further support the interpretation that attribute frequency captures heterogeneous forms of salience rather than simple review length. The findings further indicate that attribute fre quency and sentiment do not operate uniformly across service dimensions or hotel tiers. By integrating sentence-level attribute construction with multilevel modeling, this study distinguishes robust average satisfaction drivers from context- dependent service attributes and offers practical implications for context-sensitive service management in hotels.