Aspect-Based Sentiment Analysis of Multilingual Hotel Reviews in Jakarta Using Multilingual Transformer Models: A Comparative Study of Multilingual BERT and Cross-Lingual RoBERTa

Online reviews have become a critical source of information for evaluating service quality in the hospitality industry.However, extracting fine-grained insights from multilingual user-generated content remains challenging due to linguistic variability and the presence of code-mixed expressions.Aspect-based sentiment analysis (ABSA) provides an effective framework for identifying customer opinions toward specific service attributes.This study investigates the effectiveness of multilingual transformer models for ABSA on online reviews of five-star hotels in Jakarta.A large-scale dataset comprising more than 96,000 reviews collected from TripAdvisor and Google Reviews was analyzed.The proposed framework adopts a sentence-pair classification strategy that reformulates ABSA as a natural language inference task, enabling transformer models to capture aspect-sentiment relationships more effectively.Two multilingual pretrained language models-multilingual Bidirectional Encoder Representations from Transformers (mBERT) and Cross-lingual Language Model -Robustly Optimized BERT Pretraining Approach (XLM-RoBERTa)were fine-tuned and systematically compared.Experimental results show that XLM-RoBERTa achieved the best performance with an accuracy of 97.20% and an F1-score of 0.9729, slightly outperforming mBERT while requiring higher computational resources.In contrast, mBERT demonstrated greater stability across validation folds.Aspect-level sentiment analysis further revealed that cleanliness, facilities, and service are the most positively perceived aspects of five-star hotels in Jakarta, while pricing remains the primary source of negative sentiment.These findings demonstrate the effectiveness of transformerbased ABSA for multilingual hospitality reviews and provide actionable insights for datadriven decision-making in the luxury hotel sector.