Group Recommendation Model Based On Attention Mechanism And Group Member Difference

Recommender system is a powerful tool to solve information overload.People often participate in activities in the form of groups,so group recommender system comes into being.The existing group recommendation methods focus on finding the similarity between group members,but ignore the differences between members,which will greatly reduce the satisfaction of group members with the recommendation results.In addition,the aggregation strategy used is static and simple,which makes it difficult to flexibly capture the dynamic weights of group members.Therefore,it is necessary to solve the problems of how to analyze the differences between group members,how to obtain and analyze the dynamic weights of group members,so as to obtain better group recommendation results.The main contributions of this paper are as follows:(1)Aiming at the problem of low recommendation accuracy in the case of large group difference in group discovery,this paper proposes an Adaptive grouping algorithm based on user trust value(GUT).The GUT grouping algorithm mainly consists of two steps.Firstly,the K-means algorithm is used to preprocess the historical group activity information to obtain the potential groups.Then,through the user difference calculation method with user trust value,the potential group was divided into different groups and similar groups.(2)In order to solve the problem that most of the predefined strategies in group recommendation are not flexible enough to aggregate user preferences in the group,this paper fuses the Attention mechanism based group recommendation algorithm(AGREE)and Group Recommendation Algorithm Focusing on user Differences(UDA),and proposes a combined algorithm: Group recommendation algorithm based on Attention mechanism and member Difference(GAMMD).The algorithm uses two multi-layer perceptrons with attention mechanisms to learn the weights of group members and aggregate the preferences of group members respectively.The AGREE algorithm is used for the recommendation of similar groups,and the UDA algorithm is used for the recommendation of dissimilar groups.(3)Compared with the baseline method on the CAMRa2011 and Amazon Review Data(2018)datasets,the proposed GAMMD model has more than 6.4% improvement in HR and NDCG.In addition,a movie recommendation prototype system based on the algorithm is constructed to verify the feasibility of the proposed method.