Research On Recommendation Algorithm Based On Attention Mechanism

With the growth of users and the rapid development of the internet,the collection and dissemination of information have reached an unprecedented level.Information is pouring into all aspects of people's lives in waves.If users cannot reasonably use information and analyze valuable information,the information obtained is junk information or false information..User information consumption needs cannot be met.This is the "information overload" problem.The effective way to solve information overload is the recommendation system.Recommendation systems are divided into personalized recommendation systems and group recommendation systems according to different organizational structures.The personalized recommendation system only considers the needs of a single user and recommends a single user's favorite items.The group recommendation system predicts the correlation between this group and the project.Try to maximize the recommendation results to meet the needs of all members and minimize the conflict of preferences among team members.This article mainly studies from these two directions?1?In the personalized recommendation system,the recommendation algorithm based on the generated adversarial network model only uses the user-item rating matrix information,and does not take into account the user's social attribute information and item label information.Therefore,this paper proposes a recommendation algorithm(IRGAN-Deep FM)that fuses IRGAN and Deep FM.The advantages of the Deep FM model can extract low-order features and high-order combined features at the same time,without artificial feature engineering.The generative model and discriminant model of generative adversarial networks are learned in the game.The algorithm improves the identification accuracy of the discriminant model and generates more realistic sample data.The experimental results show that the proposed method has better recommendation prediction accuracy and ranking quality than the comparative experiments.2? In the group recommendation system,the recommendation algorithm generally has the problems of matrix sparseness and group preference fusion.This paper presents a bayesian group recommendation algorithm based on attentional mechanism(ANBGR).The algorithm uses the interaction record of the group and the project to generate a triple of groups,positive feedback items,and negative feedback items.Then we take advantage of the attention model to consider two computing strategies: user preference aggregation and preference game between subgroups.The weights of users and subgroups are calculated dynamically.we use a multi-layer perceptron network structure to fit the nonlinear relationship between the user and the project.Finally,we use Bayesian theory to realize the group's preference prediction of the project.