Sentiment Analysis Impact on Hotel Occupancy Prediction

Reviews have been shown to strongly influence consumer behavior in the hospitality industry and meanwhile became a powerful tool for accommodation managers. However, this rich data source still represents an underutilized asset due to the lack of a deepened understanding of the behavior of readers of such types of feedback. Whereby guest sentiment specifications in forecasting models are theoretically justified, they lack profound empirical evidence to empower their evaluation. This paper solves the not yet optimized integration of guest sentiment inputs to enhance the accuracy of occupancy (OCC) demand forecasting models, here ARIMA(X) models, at the single resort level. BERT-based textual classifications are combined with numerical review ratings expressed online to generate monthly perceptions in a hybrid sentiment analysis approach. Their influence on OCC rates is weighed in a simulation study to reveal the optimized ratio between the two forms of guest feedback. Furthermore, the temporal decay of the relevance of guest feedback is modeled in a data-driven way. This study expands our knowledge about (1) the importance ratio between textual and numerical guest feedback, and (2) the time decay in the relevance of preceding reviews from a temporal perspective.