Analysis of the Properties that Affect the Accuracy of a Group Recommender System

Abstract

Group recommender systems are usually built around a property that characterizes the groups (e.g., the size or the cohesion). However, the performance a system always measures how accurate the produced recommendations are and no study shows if the properties that characterize a group have an impact on the accuracy of the system (e.g., if more cohesive groups lead to more accurate recommendations). This paper presents a novel study of the correlation between the properties that characterize a group and the accuracy of the system for that group. This local analysis helps understanding which properties of a group have an impact on the accuracy. Thanks to this study, the design of a group recommender systems can be improved, by tailoring the recommendations on the characteristics of the groups. Experimental results show that the properties that affect the performance of a system are those related to the cohesiveness of a group.