Recommending Hotels Based On Multi-Dimensional Customer Ratings

Recommender Systems (RS) have shown to be a valuable means to support the traveller or tourist in his pre-trip information search and decision making processes. These systems often rely on rating information provided by the user community to make recommendations for individual users. In classical application domains such as movie or book recommendation, users provide one overall rating for each item. Customers in the travel and tourism domain however are often allowed to evaluate their hotel or holiday packages along several dimensions after the trip. In this work, we show through an empirical evaluation based on a real-world data set from the tourism domain that the predictive accuracy of an RS can be significantly improved when the multi-dimensional rating information is taken into account. In particular, we demonstrate that regression-based methods and in particular the novel combination of user- and item-based models leads to more accurate recommendations than previous approaches. In addition, we show that not all dimension (criteria) ratings are equally valuable for the prediction process and that a careful selection of rating dimensions can help to further increase the quality of the recommendations.