A Decision-Based Approach for Recommending in Hierarchical Domains L.M. de Campos, J.M. Fern´ andez-Luna, M. G´omez, and J.F. Huete Departamento de Ciencias de la Computaci´on e Inteligencia Artificial, E.T.S.I. Inform´atica, Universidad de Granada, 18071 – Granada, Spain {lci, jmfluna, mgomez, jhg}@decsai.ugr.es Abstract. Recommendation Systems are tools designed to help users to find items within a given domain, according to their own preferences expressed by means of a user profile. A general model for recommen- dation systems based on probabilistic graphical models is proposed in this paper. It is designed to deal with hierarchical domains, where the items can be grouped in a hierarchy, each item being only contained in another, more general item. The model makes decisions about which items in the hierarchy are more useful for the user, and carries out the necessary computations in a very efficient way. 1 Introduction In this paper we present an approach to recommending in hierarchical domains that poses this problem as a decision-based task. Broadly speaking, a Recom- mendation System (RS) provides specific suggestions about items or actions, within a given domain, that may be considered interesting to the user [11]. The input of a RS is normally expressed by means of information given by the user about his/her tastes or preferences, provided either explicitly (by means of a form or a questionnaire) or implicitly (using purchase records, viewing or rating items, visiting links, taking into account the membership to a certain group,...). All the information about the user that the RS stores is known as the user profile. The main characteristic of RSs is that they do not only return the requested information, but also try to anticipate user’s needs. There are two main types of RSs: Content-based and Collaborative filtering RSs. The former tries to recommend items based exclusively on the user prefer- ences, whereas the latter tries to identify groups of people with tastes similar to that of the user and recommends items that they have liked [1]. A much more exhaustive classification of RSs is found in [8]. In order to place the problem as a decision task we shall use the probabilistic graphical models formalism. Different approaches to the RS are found in the literature: One of these are Bayesian networks (BN) that have been used in this field basically in two areas: as the tool on which the user profile is built[14, 10, 15, 3] and collaborative filtering, employed in classification tasks [2, 9, 12]. L. Godo (Ed.): ECSQARU 2005, LNAI 3571, pp. 123–135, 2005. c  Springer-Verlag Berlin Heidelberg 2005