Online reviews significantly shape public perception and play a crucial role in customer decision-making within the hospitality sector. This research aims to conduct aspect-based sentiment analysis on Indonesian five-star hotel reviews using a fine-tuned IndoBERT model. Unlike prior studies that mainly applied IndoBERT to single hotels or small-scale datasets, this study fills that gap by examining 2,499 reviews collected from five luxury hotels in Jakarta. The analysis focuses on five essential service aspects: cleanliness, service quality, room comfort, food & beverages, and core facilities. The IndoBERT-base model was fine-tuned with annotated aspect-sentiment data and assessed using accuracy, precision, recall, F1-score, and confusion matrices. Experimental results show that the model reached 95.28% accuracy with a macro F1-score of 82.44%. Positive sentiment dominated the reviews (81.4%), while neutral and negative sentiments represented 16.9% and 1.7%, respectively. Service, along with food & beverages, received the highest praise, whereas cleanliness and core facilities were more often evaluated neutrally. Aspect and sentiment annotations were carried out semi-automatically using large language models (LLMs) and later validated by human annotators to ensure reliability. These findings highlight IndoBERT’s strong capability in aspect-based sentiment classification for Indonesian hotel reviews and provide actionable insights for hotel managers to enhance service quality. Moreover, this study demonstrates both the academic and practical significance of applying fine-tuned Transformer models to real-world customer experience analysis.