The Application Of Belief Propagation In Group Recommendation Algorithm

Recommendation system is an important method of information retrieval that retrieves interesting items for users based on the past activities and user's profile.With the popularity of e-commerce,the recommendation system has made great development,especially in Amazon and Netflix make the recommendation system has received great attention.The current recommendation system is mainly aimed at individual users' recommendation.However,many of our daily activities are performed by a group of multiple users,such as friends watching movies and traveling together.In this case,both project dependencies and disagreements among the members need to be considered,and the recommendation system of individual users may not be able to meet such needs,and the group recommendation has come into being.In recent years,some group recommendation algorithms have been proposed one after another,most of which are collaborative filtering algorithms based on aggregation strategies.Therefore,it is inevitable to face the data sparseness of collaborative filtering algorithm and the cold start of new project.In addition,the current recommendation system mainly focuses on user's rating of a project,that is,recommending a project to a group or individual by predicting a group or individual's rating on the project,in some cases,such as online shopping,a ranked list of items Favor by the group,the ranking of the project at this time is more important than the simple prediction score.To solve the above problems,this paper proposes a TOP-N group recommendation algorithm based on BP algorithm.Firstly,a certain preference fusion strategy is adopted to calculate the score of the whole group according to the score of the group members.Then,the entire group is taken as a single user,and the group set and the item set are set as two node sets The bipartite graph,the group score of the project as the edge threshold of the group nodeand the project node,implicitly mining the group's contribution to the project by passing the confidence between the project group nodes in the Markov field Preference,according to the final goal of the confidence level generated by the node to generate ranking list recommended to the group.Taking into account the Belief Propagation algorithm itself using the global node,there is a greater redundancy calculation,this article without loss of accuracy under the premise of adaptive size nodes to calculate node confidence,that is,adaptive Belief Propagation algorithm.Finally,we experiment with the algorithm on the Open Source MovieLens dataset provided by the GroupLeans project.Experimental results show that the group recommendation algorithm based on confidence propagation algorithm proposed in this paper has higher nDCG value and higher recommendation accuracy than other TOP-N recommendation methods.At the same time,the group recommendation algorithm based on the adaptive belief propagation algorithm has higher computational speed than the group recommendation algorithm based on the confidence propagation algorithm when the accuracy rate is guaranteed.