Unpacking the Emotional Landscape of Reviews: Sentiment-Augmented Topic Modeling with Transformer Embeddings

Abstract - This research introduces a novel methodology for analysing Trip Advisor reviews by integrating sentiment analysis directly into the feature engineering stage of transformer-based topic modeling (BERTopic). Moving beyond traditional sequential approaches, this method simultaneously captures thematic content and associated sentiment, providing a nuanced understanding of customer perceptions. We identify 26 distinct topics and analyze their prevalence across different star rating classes. Results reveal strong correlations between specific topics (e.g., staff and cleanliness) and positive reviews, while others (e.g., negative experiences) dominate lower ratings. Quantitative evaluation using coherence scores indicates meaningful semantic relationships within the discovered topics. This integrated approach offers a richer, more context-aware understanding of customer feedback, enabling hotels to pinpoint key drivers of satisfaction and dissatisfaction. The study demonstrates the efficacy of sentiment-augmented topic modeling in extracting actionable insights from online reviews, offering a valuable framework for hospitality research and industry applications. KEYWORDS: Topic Modeling, Sentiment Analysis, Transformer Embeddings, BERTopic, Hotel Reviews, Customer Perception, Text Mining, Hospitality Industry.