Group detection based on affinity propagation in group recommendation system

A group detection algorithm is proposed in group recommendation system based on affinity propagation clustering (APC), which clusters users according to the ratings predicted by the user-based collaborative filtering algorithm to generate potential groups for group recommendation. There is no need to set the number of clusters and the initial representation paradigm in the APC so that the group detection results avoid the instability problem caused by randomly selecting the initial typical sample, thereby generating more similar groups in group recommendation system. The data of the top 100 and top 200 users in the MovieLens data set is selected for testing. The experimental results of the proposed method are compared with the k-means algorithm and it is found that the indicators MAE^G and D^G for assessing group consistency and satisfaction are decreased by 0.14 and 0.02. In ongoing work, a group recommendation system is being planned by extracting more user characteristics and item features to detect groups.