V. Wade, H. Ashman, and B. Smyth (Eds.): AH 2006, LNCS 4018, pp. 21 – 30, 2006. © Springer-Verlag Berlin Heidelberg 2006 Cross-Technique Mediation of User Models Shlomo Berkovsky 1 , Tsvi Kuflik 1 , and Francesco Ricci 2 1 University of Haifa, Haifa slavax@cs.haifa.ac.il, tsvikak@is.haifa.ac.il 2 ITC-irst, Trento ricci@itc.it Abstract. Nowadays, personalization is considered a powerful approach for de- signing more precise and easy to use information search and recommendation tools. Since the quality of the personalization provided depends on the accuracy of the user models (UMs) managed by the system, it would be beneficial en- riching these models through mediating partial UMs, built by other services. This paper proposes a cross-technique mediation of the UMs from collaborative to content-based services. According to this approach, content-based recom- mendations are built for the target users having no content-based user model, knowing his collaborative-based user model only. Experimental evaluation conducted in the domain of movies, shows that for small UMs, the personaliza- tion provided using the mediated content-based UMs outperforms the personal- ization provided using the original collaborative UMs. 1 Introduction The quantity of information available on the Web grows rapidly and exceeds our limited processing capabilities. As a result, there is a pressing need for intelligent systems providing personalized services according to user's needs and interests, and delivering tailored information in a way most appropriate to the user [10]. Providing personalized services to the users requires modeling their preferences, interests and needs. This data is referred in the literature as a User Model (UM) [8]. Typically, service providers build and maintain proprietary UMs, tailored to the application domain of the service and to the specific personalization technique being exploited. Since the accuracy of the provided personalized service heavily depends on the characteristics and quality of the UMs, different services would benefit from en- riching their UMs through importing, translating and aggregating partial UMs, i.e., UMs built by other, possibly related, services. This can be achieved through media- tion of partial UMs [2]. The main functionality of UM mediator [2] is to acquire partial UMs built by other service providers, and to aggregate the acquired UMs into a UM for the target service. Analysis of the state-of-the-art personalization techniques and application domains yields four groups of services that can potentially provide valuable partial UMs for building a UM for a service from domain d exploiting technique t: (1) services from d that also exploit t, (2) services from d that exploit another technique t', (3) services