Research On Recommendation Of Graph Neural Networks Based On Social Trust Networks

The development of the Internet has brought convenience to people while also exposing them to the problem of information overload.As an effective solution to information overload,recommendation systems can select useful information for users according to their preferences,and their core is to generate personalized recommendation results for users by designing recommendation algorithms.In recent years,graph neural networks have shown powerful capabilities in processing nonEuclidean data and learning feature representations of graph nodes,and have played an important role in the research on recommendation,especially the recommendation of social trust networks has been introduced.However,the existing research on recommendation based on graph neural networks ignores the transfer characteristics of social trust networks,and there are problems such as insufficient consideration of the implicit relationship information between graph nodes and poor recommendation performance.In order to solve the above problems,this paper proposes a graph neural network recommendation model named TGNRec that applies the transmission characteristics of social trust,and a graph neural network recommendation model named Trust＿TGNRec that based on trust values.The works and innovation points of this paper are as follows:(1)In response to existing studies that do not fully consider the influence of users’ indirect trust on users’ preferences when using social trust networks,this paper applies the transmission of social trust networks and proposes the graph neural network recommendation model TGNRec,which considers both direct and indirect trust item interaction features of user nodes when applying graph neural networks to learn the latent feature representation of user nodes.(2)To address the problem of insufficient utilization of implicit information among graph nodes in existing studies,this paper model TGNRec makes full use of the user-item rating interaction data when applying graph neural networks to learn the latent feature representations of item nodes,calculates the rating mean of each item and divides the similarity level based on the rating mean of each item,and establishes similar relationships among items with similar ratings.In addition,this paper uses the latent feature representations of user nodes and item nodes learned by the model TGNRec to achieve the prediction of rating of items by target users.(3)Although the model TGNRec demonstrates that the application of social trust transfer and the implicit relationship of items can improve the recommendation performance,the trust relationships in the social trust networks have the problem of loss in the transfer process and the model TGNRec is deficient in establishing the item similarity relationship based on the rating mean.To address these problems,we propose Trust-U,a user trust value calculation method,and P-S,an item relationships graph building algorithm based on Pearson correlation coefficient.And applies TrustU and P-S to proposes Trust＿TGNRec,a graph neural network recommendation model based on trust values.The model first combines Trust-U,a user trust value calculation method,with an attention mechanism for learn the potential feature representation of user nodes,and then use the algorithm P-S to obtain the similar relationship graph of items and combine the user-item rating interaction to learn the potential feature representation of item nodes.Finally,this paper conducts performance validation experiments for both models on two real-world datasets.Compared with other recommendation models,the above models can reduce the error of rating prediction and improve the recommendation performance.