Meaningful hotel recommendation based on hotel review text

In the digital travel ecosystem, user reviews significantly influence hotel choices. However, most recommendation still rely mainly on aggregated numerical ratings, often overlooking the deeper insights in review text. This thesis uses Aspect-Based Sentiment Analysis (ABSA) to extract detailed sentiment information from reviews and combines it with numerical ratings to develop a more accurate and fair hotel recommendations. This research investigates hotel guest satisfaction using 134,550 multilingual reviews from Booking.com for hotels in Singapore. Leveraging ChatGPT o3-mini, sentiment scores were extracted for seven key aspects, including comfort, cleanliness, and location. These were integrated with numerical ratings through Principal Component Analysis and modelled using a Cumulative Link Model (CLM) to predict ordinal hotel ratings. The model, validated on 24,172 instances, demonstrated high predictive performance, with over 92% of predictions matching actual ratings exactly. Statistical analysis revealed that sentiments related to comfort, facilities, and cleanliness significantly increased satisfaction, while location sentiment had a negative impact. These findings highlight the value of review text in capturing guest experiences often overlooked by traditional rating systems. The proposed hybrid model offers a more personalized and accurate recommendation approach and provides actionable insights for hotel management seeking to improve service quality and guest satisfaction. This research presents a validated framework for building objective, context-aware hotel recommendation systems that consider both guest ratings and the reasons behind them. The proposed method, backed by strong empirical results, offers a reliable tool for booking platforms and hotel managers to enhance guest-hotel matching and improve satisfaction through meaningful analysis of review texts.