User Modeling for Personalized City Tours 1 Josef Fink humanIT – Human Information Technologies Sankt Augustin, Germany Josef.Fink@humanIT.com Alfred Kobsa Dept. of Information and Computer Science University of California, Irvine, U.S.A. kobsa@uci.edu Abstract Several current support systems for travel and tourism are aimed at providing informa- tion in a personalized manner, taking users interests and preferences into account. In this vein, personalized systems observe users behavior and, based thereon, make gener- alizations and predictions about them. This article describes a user modeling server that offers services to personalized systems with regard to the analysis of user actions, the representation of assumptions about the user, and the inference of additional assump- tions based on domain knowledge and characteristics of similar users. The system is open and compliant with major standards, allowing it to be easily accessed by clients that need personalization services. Keywords : Personalization, user modeling server, learning about the user, interest profile, LDAP, mobile tourist guide 1 Introduction Computer support for travel and tourism has recently attracted considerable interest, both with regard to research and experimental deployment. Possible assistance for travelers and tourists ranges from web-based travel planning to stationary information kiosks and then on to location- aware portable museum and city guides. Since tourism is intimately connected with personal interests and preferences, many of the systems developed in this area aim at providing informa- tion in a personalized manner [Abowd et al. 1997; Fink et al. 1998; ILEX 1998; Not et al. 1998; Cheverst et al. 2000a; Malaka and Zipf 2000; Oppermann and Specht 2000; Poslad et al. 2001]. Personalization means that systems cater to each individual user, thereby taking e.g. his interests, preferences and background knowledge into account. As a prerequisite, personalized systems must be able to watch the users behavior and make generalizations and predictions about the user based on their observations. This information about him is usually collected in a so-called user model and administered by a user modeling system [Wahlster and Kobsa 1989; Kobsa et al. 2001]. The work presented here was carried out in the context of the Deep Map project [Malaka and Zipf 2000; Deep Map 2001] of the European Media Laboratory in Heidelberg, Germany. It also benefited from prior experience in the AVANTI project [Fink et al. 1998]. Deep Map is part 1 Part of the research presented here was supported by a grant of the European Media Lab to GMD  German National Research Center for Information Technology. Artificial Intelligence Review 18, 2002, pp. 33-74, Kluwer Academic Publishers.