The exponential rise of online review platforms has turned hotel reviews into a rich source of information related to customer experiences. While most of these platforms provide numerical ratings for capturing broad assessments of satisfaction across various service dimensions such as location or room quality, they often fail to capture the nuanced and subjective feedback expressed in textual reviews. This study introduces a novel, customer-centric clustering approach that leverages aspect–sentiment-aware semantic sentence embeddings to group hotels based on experiential similarity across multiple service dimensions. Using OpenAI’s GPT model, initially hotel reviews were classified into predefined aspects and each aspect-specific sentence was assigned a sentiment polarity (positive/negative/neutral). Then, Sentence-BERT (SBERT) was employed for generating dense semantic representations of these aspect-level review sentences. To reflect polarity, we construct sentiment-weighted aspect embeddings by assigning weights to sentences based on their sentiment before aggregation and normalization. Cosine similarity was applied to the resulting embeddings for computing pairwise hotel similarity scores. Then, these similarity scores were used for aggregating and clustering to uncover latent guest experience profiles that distinguish not only what guests talk about, but also how they feel about each aspect. The clustering results were validated through TripAdvisor aspect ratings. The proposed method offers a more nuanced framework for hotel comparison, facilitating personalized recommendations for consumers and enabling competitive benchmarking for service providers, thus advancing the application of natural language processing and aspect-based sentiment analysis in hospitality analytics.
Öztürk et al. (Tue,) studied this question.