This paper introduces a high-precision sentiment analysis system that correctly classifies TripAdvisor hotel reviews on a 5-point scale from strong negative to strong positive. In the hospitality sector, knowing customer sentiment with subtle accuracy yields vital business intelligence that binary classification models cannot obtain. Our approach is novel in that it leverages an ensemble of multiple different transformer models to achieve improved accuracy at 5-class sentiment classification. We build an ensemble model utilizing and comparing seven advanced transformer models—BERT, DeBERTa, ModernBERT, DistilBERT, RoBERTa, T5, and a self-implemented LSTM—to make sentiment predictions with finer granularity than standard positive/negative labels. Our data set is 8,000 English TripAdvisor reviews (2019–2022), professionally annotated on a 5-point sentiment scale. Preprocessing involves tokenization, normalization, and processing of sentiment-carrying punctuation. When comparing single models, the highest single accuracy is 69.66% by T5, and DeBERTa stands out in terms of F1 score at 68.53%. Nevertheless, our weighted majority voting ensemble excluding LSTM far surpasses all the single models with 70.11% accuracy and 70.20% F1 score, and boasts a remarkable 97.95% relaxed accuracy (predictions with ±1 deviation from the correct sentiment). This work shows that ensemble transformer architectures can effectively capture subtle semantic nuances in hospitality reviews, offering a robust framework for fine-grained sentiment classification that enables more granular customer experience management.