Uncovering customer insights from online hotel reviews in Vietnam: A bilingual NLP study using sentiment analysis and topic modelling

Online customer reviews are a critical source of insight for evaluating hotel service quality in the current digital era. This study analyzes over 150,000 English and Vietnamese reviews from TripAdvisor, Booking.com, and Agoda, using a multilingual NLP framework. DistilBERT and PhoBERT were employed for sentiment classification, while BERTopic extracted latent themes. The datasets show a strong positivity bias. A combined RandomOverSampler and K-fold strategy was used to address class imbalance and improve sentiment prediction. The results indicate that BERTopic outperforms LDA in coherence and interpretability, producing fine-grained topics related to staff service, cleanliness, amenities, food quality, and value for money. Cross-language comparisons reveal distinct patterns in how English and Vietnamese guests articulate negative sentiment, emphasizing the need for culturally sensitive analysis. The study provides a scalable method for multilingual hotel review analytics and offers actionable insights to further enhance service in Vietnam's hospitality industry.