Research On Social Network Recommendation Algorithm Based On Deep Learning

Due to the problem of information overload,users cannot obtain useful information accurately and efficiently.Effective mitigation of information overload requires the use of recommender systems.The recommendation system can learn user preferences based on the user’s attribute information and browsing and purchase interaction records,and efficiently and accurately push content for users,saving users time.At the same time,the recommendation system also increases user stickiness and economic benefits for the Internet platform.Recommendation system is indispensable on the Internet platform.The recommendation system has the problem of data sparsity and cold start.How to solve the above problems is the main goal of the related research of the recommendation system.User’s social network can reflect the social connections between users and friends.In recent years,academic circles have paid extensive attention to recommendation algorithms based on social networks.Introducing social networks into the recommendation system can effectively utilize user friend information to alleviate the data sparsity problem of the user-item interaction matrix,so that the recommendation system can better capture user preferences and improve recommendation performance.Graph neural networks have powerful mining capabilities for non-European spatial data such as social networks,and are widely used in social network recommendation research.Aiming at the above problems,this paper uses social network,uses graph neural network to perform representation learning on the social network graph and user-item interaction graph used in this paper,and proposes two social recommendation algorithms based on graph neural network: Graph Convolutional Network Based Social Recommendation(GBSR model)and the Graph Attention network Based Social Recommendation(GABSR model)that integrates social networks,the work done is as follows:First,aiming at the data sparsity problem in the recommendation system,this paper uses user social relationship data and builds a user social network graph.The GBSR model is proposed.First,the GBSR model can process the user social network network and the user-item interaction graph separately through the graph convolutional network,and learn the feature representations of users and items from multiple perspectives.The GBSR model can obtain high-level aggregation information of nodes,and through multi-layer message transmission,dig deep-level information in the graph structure.Compared with other recommendation algorithms,the GBSR model shows a better recommendation effect.Second,on the basis of the GBSR model,this paper further improves the GBSR model and proposes the GABSR model.The GABSR model assigns different attention weights to different nodes through the attention mechanism during the node aggregation process,and aggregates neighbor nodes according to the weights.information,which can further efficiently and accurately learn the embedding representation of nodes.Compared with the GBSR model,the GABSR model further improves the recommendation performance metrics.