Adaptive Personalized Travel Information Systems: A Bayesian Method to Learn Users´ Personal Preferences in Multimodal Transport Networks

Providing personalized advice is an important objective in the development of advanced traveler information systems. In this paper, a Bayesian method to incorporate learning of users´ personal travel preferences in a multimodal routing system is proposed. The system learns preference parameters incrementally based on travel choices a user makes. Existing Bayesian inference methods require too much computation time for the learning problem that we are dealing with here. Therefore, an approximation method is developed, which is based on sequential processing of preference parameters and systematic sampling of the parameter space. The data of repetitive travel choices of a representative sample of individuals are used to test the system. The results indicate that the system rapidly adapts to a user and learns his or her preferences effectively. The efficiency of the algorithm allows the system to handle realistically sized learning problems with short response times even when many users are to be simultaneously processed. It is therefore concluded that the approach is feasible; problems for future research are identified.