Socially-Aware Recommendation for Over-Constrained Problems

Abstract

Group recommender systems support the identification of items that best fit the individual preferences of all group members. A group recommendation can be determined on the basis of aggregation functions. However, to some extent it is still unclear which aggregation function is most suitable for predicting an item to a group. In this paper, we analyze different preference aggregation functions with regard to their prediction quality. We found out that consensus-based aggregation functions (e.g., Average, Minimal Group Distance, Multiplicative, Ensemble Voting) which consider all group members’ preferences lead to a better prediction quality compared to borderline aggregation functions, such as Least Misery and Most Pleasure which solely focus on preferences of some individual group members.