Spatially Embedded Experience Economy: Machine Learning-Enhanced Analysis of Peer-to-Peer Accommodation Satisfaction Beyond Price Paradigms
摘要
This study reconceptualizes guest satisfaction in peer-to-peer accommodation as a spatially embedded experiential outcome. Combining transformer-based NLP (Sentence-BERT, BERTopic) with multilevel and spatial econometric models, we analyze 109,940 Airbnb reviews from 1,092 Hong Kong listings (2011–2023) using strict temporal holdout validation. Experiential quality emerges as the dominant determinant of guest satisfaction, exhibiting a large unique explanatory contribution (partial R² ≈ 0.78), while amenities play a complementary role (partial R² ≈ 0.57). In contrast, economic factors such as price contribute only marginally once experiential and amenity dimensions are accounted for (partial R² ≈ 0.01). A Shapley R² decomposition corroborates this pattern, attributing approximately 70% of the total explained variance to experiential factors. Spatial Durbin Models further show significant spatial clustering and spillovers in satisfaction, with price effects becoming weak or non-significant after accounting for spatial dependence. Overall, the findings suggest that guest satisfaction in peer-to-peer accommodation markets is driven primarily by experiential quality rather than price competition, highlighting the importance of experience-oriented and place-sensitive strategies for platform governance and urban tourism policy.