<![CDATA[The increasing volume of hotel information on online travel platforms made hotel selection more difficult for users because decision making had to consider multiple aspects simultaneously, including value, accessibility, service, room quality, cleanliness, and sleep quality. Conventional recommendation methods often depended on overall ratings and therefore were not sufficiently capable of representing the multidimensional nature of hotel preferences. This study proposed an improved multi-criteria neural collaborative filtering (MCNCF) model for hotel recommendation using the Bali Hotel Review dataset. The proposed model integrated user identity, hotel identity, and six structured hotel evaluation criteria to learn user preferences in a more detailed and preference-sensitive manner. The experimental design was also strengthened through a more reliable preprocessing and evaluation pipeline, including data splitting before scaling, training-based imputation for missing values, and user ranking evaluation. The model was implemented using embedding-based neural interaction learning to capture nonlinear relationships between users, hotels, and multi-criteria features. The results showed that the proposed approach achieved stable and competitive performance across testing splits of 10%, 20%, 30%, and 40%. On the original rating scale, the model produced the best Root Mean Square Error of 0.416400 and the lowest Mean Absolute Error of 0.351719. In addition, the ranking performance remained consistently high, with Normalized Discounted Cumulative Gain values ranging from 0.976540 to 0.996243. These findings demonstrated that the proposed approach provided an effective and robust solution for hotel recommendation by leveraging structured multi-criteria preference information within a neural recommendation framework.]]>
Copyrights © 2026
