Article Preview
Top1. Introduction
Group recommender system (GRS) becomes a crucial tool to develop an information system supplying recommendation to joined activity of group (Dara et al., 2020). GRSs can be categorized into two common approaches in which the former creates a group profile by merging those of all members and uses it as a pseudo single user in the process of recommendation (Da’u & Salim, 2020). The later called consensus phase generates all group members’ preferences and aggregates them to select the suitable recommendation to group (Kuhlman & Rundensteiner, 2020; Banda et al., 2020).
One of the most concerned problems in GRSs is to guarantee the fairness between users in a group so-called the consensus-driven group recommender system. This approach makes recommendation for the group based on individual interests using consensus mechanisms. It utilizes the information of individuals, and a set of preferred items recommended to a group based on individuals’ sets of preferred items. The fairness issue in GRSis a social concept and not easy to measure(Kaya, Bridge & Tintarev, 2020). Recently, several researchers have defined the fairness in GRS explicitly as the ratio of satisfied people to total group members (Felfernig et al., 2018), the equity of group members satisfactions (Xiao et al., 2017) or consider fairness of recommended item set as a package rather than a set of independent items (Serbos et al., 2017). Nevertheless, a challenge raised in finding good fair solution in consensus-driven GRS is that a member preference for an item is often influenced by members interaction and item relations (Wang et al., 2017). Therefore, it is difficult to estimate the imbalance between group members’ preferences when regarding members’ interactions.
When dealing fairness issue, the common GRSs do not deal with effect of interaction between group’s members. In this case, the user-item relevance do not change when group members change, thus the fairness of recommendation is simply calculated from individual preferences (Dara et al., 2020). On the other hand, a number of studies presented solutions for estimating group preference regarding group member relationship from external information such as the user social network profile (Yin et al., 2020) or internal information from the group member distances (Castro et al., 2015). However, these studies did not stress on the fairness problem in GRSs as expected.
In order to cope the fairness issue, we are motivated by the aggregation operators in the consensus phase of GRS. That is to say, if a well-defined aggregation operator is formed, we can achieve the fairness between users in a group while still maintaining reasonable accuracy of prediction and recommendation. There are many aggregation operators presented in the literature to generate group preference such as the additive utilitarian strategy, average strategy, least misery strategy, approval voting strategy, fairness strategy, and so on (Cantador & Castells, 2012). Several strategies maximize the total group preference such as additive utilitarian strategy, average strategy and other strategies promote weaken member such as least misery strategy, Borda Count strategy, approval voting strategy, fairness strategy. However, these strategies do not solve fairness issue of group RS directly, even fairness strategy does not deal with the imbalance of satisfactions between group members.