L. M. de Campos et al. group members to collaboratively specify their preferences and to reach an agree ment about an overall solution. In this case, a group profile is obtained through the nteraction of the members, taking into account the systems current recommenda tion which is obtained by aggregating individual preferences for each dimension In the article(Kudenko et al. 2003), a system is presented to help a group of users reach a joint decision based on individual user preferences CB+AR+PM: Intrigue(Ardissono et al. 2003)recommends tourist attractions for heterogeneous groups that include homogeneous subgroups where the members have similar preferences. In this system, the users record their preferences for a series of tourist attractions, and recommendations(obtained using a fuzzy AND) are then merged using a weighted scheme where each weight represents the rele vance of the corresponding subgroup(for instance, a subgroup could be particularly influential since it represents a large portion of the group) explains their recommendations, it has no means of interacting with the user CF+AR+PM: Polylens(O'Connor et al. 2001), an extension of MovieLens(Her locker et al. 2004), recommends movies to groups of users. This system uses a nearest neighbor-based algorithm to find the individuals with the most similar tastes to those of each group member and to obtain recommendations for every user. The voting preferences of these individuals are then merged according to the principle of least misery(minimum criterion). Under the same classification( Chen et al 2008)uses genetic algorithms to learn the group rating for an item that best fits the existing ratings for the item given by the individuals and the subgroups. The idea is that it is possible to learn how the user interacts from the known group ratings The proposed algorithm therefore recommends items based on the group's previous ratings for similar items. 2.1.1 The role of uncertainty As far as the authors are aware, the role of uncertainty in group recommending pro- cesses has not been considered. Nevertheless, many papers have been published which tackle this problem when recommendations are made to individual users(Zuker man and Albrecht 2001; Albrecht and Zukerman 2007). Focusing on probabilistic approaches, those relating to the one presented in this paper include content-based RSs(Mooney and Roy 2000: de Campos et al. 2005), collaborative filterin (Breese et al. 1998; Schiaffino and Amandi 2000: Butz 2002; Lekakos and 2007: Miyahara and Pazzani 2000: Heckerman et al. 2001)and hybrid methods(Pope- scu et al. 2001: de Campos et al. 2006). In terms of the group's process, the treatment of uncertainty is, however, a well known problem in other disciplines and so in this section we will review those papers which focus on the combination of probabilistic information from a purely approach(see Clemen and winkler 1999: Genest and Zidek 1986). In general, we might consider these methods as analytical models operating on the individual prob ability distributions to produce a single"combined"probability distribution. These approaches can generally be further distinguished into axiomatic approaches(by con- sidering a set of assumptions that the combination criteria might satisfy) and Bayesian approaches212 L. M. de Campos et al. group members to collaboratively specify their preferences and to reach an agree￾ment about an overall solution. In this case, a group profile is obtained through the interaction of the members, taking into account the system’s current recommenda￾tion which is obtained by aggregating individual preferences for each dimension. In the article (Kudenko et al. 2003), a system is presented to help a group of users reach a joint decision based on individual user preferences. – CB+AR+PM: Intrigue (Ardissono et al. 2003) recommends tourist attractions for heterogeneous groups that include homogeneous subgroups where the members have similar preferences. In this system, the users record their preferences for a series of tourist attractions, and recommendations (obtained using a fuzzy AND) are then merged using a weighted scheme where each weight represents the rele￾vance of the corresponding subgroup (for instance, a subgroup could be particularly influential since it represents a large portion of the group). Although the system explains their recommendations, it has no means of interacting with the user. – CF+AR+PM: Polylens (O’Connor et al. 2001), an extension of MovieLens (Her￾locker et al. 2004), recommends movies to groups of users. This system uses a nearest neighbor-based algorithm to find the individuals with the most similar tastes to those of each group member and to obtain recommendations for every user. The voting preferences of these individuals are then merged according to the principle of least misery (minimum criterion). Under the same classification, (Chen et al. 2008) uses genetic algorithms to learn the group rating for an item that best fits the existing ratings for the item given by the individuals and the subgroups. The idea is that it is possible to learn how the user interacts from the known group ratings. The proposed algorithm therefore recommends items based on the group’s previous ratings for similar items. 2.1.1 The role of uncertainty As far as the authors are aware, the role of uncertainty in group recommending pro￾cesses has not been considered. Nevertheless, many papers have been published which tackle this problem when recommendations are made to individual users (Zuker￾man and Albrecht 2001; Albrecht and Zukerman 2007). Focusing on probabilistic approaches, those relating to the one presented in this paper include content-based RSs (Mooney and Roy 2000; de Campos et al. 2005), collaborative filtering RSs (Breese et al. 1998; Schiaffino and Amandi 2000; Butz 2002; Lekakos and Giaglis 2007; Miyahara and Pazzani 2000; Heckerman et al. 2001) and hybrid methods (Pope￾scu et al. 2001; de Campos et al. 2006). In terms of the group’s process, the treatment of uncertainty is, however, a well￾known problem in other disciplines and so in this section we will review those papers which focus on the combination of probabilistic information from a purely statistical approach (see Clemen and Winkler 1999; Genest and Zidek 1986). In general, we might consider these methods as analytical models operating on the individual prob￾ability distributions to produce a single “combined” probability distribution. These approaches can generally be further distinguished into axiomatic approaches (by con￾sidering a set of assumptions that the combination criteria might satisfy) and Bayesian approaches: 123