Estimating a latent-class user model for travel recommender systems

In determining the selection of sites to visit on a trip tourists have to trade-off attraction values against routing and time-use characteristics of points of interest (POIs). For recommending optimal personalized travel plans an accurate assessment of how users make these trade-offs is important. In this paper we report the results of a study conducted to estimate a user model for travel recommender systems. The proposed model is part of c-Space—a tour-recommender system for tourists on a city trip which uses the LATUS algorithm to find personalized optimal tours. The model takes into account a multi-attribute utility function of POIs as well as dynamic needs of persons on a trip. A stated choice experiment is designed where the current need is manipulated as a context variable and activity choice alternatives are varied. A random sample of 316 individuals participated in the on-line survey. A latent-class analysis shows that significant differences exist between tourists in terms of how they make the trade-offs between the factors and respond to needs. The estimation results provide the parameters of a multi-class user model that can be used for travel recommender systems.