Research And Implementation Of Group Recommendation Algorithm Based On Self-supervised Learning And Hypergraph Representation Learning

With the rise of the Internet,people have to face tens of thousands of information every day.Recommendation system plays an important role in helping users filter redundant information.The rise of the Internet also makes the rapid development of online social networks and online communities.Users with similar interests have formed a variety of communities or groups.Online group activities are becoming more and more common.In the face of the rich and colorful information in life,the traditional personalized recommendation system is unable to meet the needs of users.Different from the traditional personalized recommendation system,which recommends to a single user,the target object of group recommendation system becomes multiple users in a group,aiming to build a recommendation system that meets the preferences of most group members.Therefore,how to integrate the preferences of group members is the difficulty of group recommendation.Previous group recommendation models mostly utilize preference fusion strategies based on predefined rules,such as mean fusion,minimum pain fusion,and maximum satisfaction fusion.However,these models ignore the different roles of members in the group,and do not consider the changes of members' preferences due to the influence of other users,which makes it impossible to accurately predict the real group preferences.In addition,in real life,the number of interactions with activities for people participating in the form of groups are very few.Although some models introduce the auxiliary information to solve the problem of data sparsity,it is not suitable for the model that cannot obtain auxiliary information.In view of the above analysis,the research on the recommendation model for occasional groups has the following difficulties:(1)Historical interaction data in occasional groups is relatively sparse,making it difficult to predict group preferences.(2)Member preferences are easily influenced by others,making it difficult to predict user preferences.(3)The preference fusion process of group members is complex and dynamic,which makes it difficult to simulate group decision-making.For these difficulties,this paper proposes a group recommendation model based on hypergraph and self-supervised learning.The main work of this paper is as follows:This paper introduces the research background,research significance,and current research status of group recommender system.By summarizing with the technology and the theory of the group recommendation,hypergraph representation learning,and self-supervised learning,the problems and shortcomings of current group recommendation model research are analyzed;This paper designs multiple hypergraph-based auxiliary tasks to achieve data enhancement,including node perturbation-based auxiliary tasks,hyperedge perturbation-based auxiliary tasks,and subgraph-based auxiliary tasks.Hypergraph neural networks are used to learn hypergraph representations under different auxiliary tasks.Then,this paper applies contrastive learning to capture the accurate characteristics of user nodes and finally utilizes the optimized node vector representation to the recommendation task to verify the effectiveness of the designed auxiliary task and lay a theoretical foundation for solving the challenge(1).The group hypergraph representation is given formally in this paper,using hyperedges to connect multiple users in the group to capture high-level interactions among users,accurately learn the interaction among users,and solve the challenge(2).This paper uses an attention-based mechanism preference fusion model to learn different weights of members in the group,simulate complex decision-making processes,and solve the challenge(3).This paper designs multi-granularity node disturbance self-supervision strategy and group-user maximizing mutual information strategy to optimize the representation of users and groups and solve the challenge(1).Through a large number of experiments based on public datasets,the superiority of the proposed algorithm is proved,and the reasons for the excellent performance of the proposed algorithm are further analyzed;According to the proposed model,a group song recommender system based on multi-granularity hypergraph self-supervised learning is designed and implemented,including group division function,group song recommendation,and model performance display to recommend songs that may be of interest to the group.