This study aims to examine the relationship between neutral sentiment and customer satisfaction in online hotel reviews by comparing binary and ternary sentiment classification models. It further explores the effects of neutral sentiment across key hotel aspects: location, cleanliness, service, value, meals and facilities.
This research includes data preprocessing, aspect classification, sentiment labeling and regression analysis. Hotel aspects are categorized using lexicons, and binary and ternary sentiment models are trained with Bidirectional Encoder Representations from Transformers on a manually labeled dataset. Model performance is evaluated using generalized ordinal logistic regression.
Incorporating neutral sentiment significantly enhances explanatory power. This study identifies distinct effects of “indifferent neutral” and “mixed neutral” sentiments on customer satisfaction and shows that their impacts vary across different hotel aspects.
This study contributes to the sentiment analysis literature by highlighting the theoretical and methodological importance of accounting for neutral sentiment and aspect-level variations.
The results suggest that hotel managers should focus on service areas with high “indifferent neutral” sentiment to improve quality and satisfaction, supporting more targeted, data-driven operational decisions.
This research offers a nuanced understanding of neutral sentiment’s role in online hotel reviews, emphasizing the value of aspect-based sentiment analysis in capturing varied guest perceptions.
