Research On Graph Neural Network Recommendation Algorithm Based On Interaction Time And Social Network

The rapid development of information technology has led to the explosive growth of data,which makes people enter an era of information overload.Nowadays,how to quickly and accurately mine useful information from massive data becomes an urgent problem.Based on this,the recommendation system came into being.Recommendation system can provide targeted information for users according to user interest preference.The core of recommendation system is the design of recommendation algorithm.Among many recommendation algorithms,collaborative filtering algorithm is undoubtedly the most classical and widely used algorithm.However,the traditional collaborative filtering recommendation algorithm usually has the problems of sparse data and cold start,and ignores the characteristics that user interest will change with time.In order to solve these problems,it is a very effective way to integrate auxiliary information.In addition,in recent years,the emergence of graph neural network makes people can use graph to dig deeper information of nodes and enrich the embedded representation of nodes.Therefore,it is a very potential research direction to use graph neural network to recommend personalized.In order to improve the performance of the algorithm and alleviate the user cold-start problem of the model,this paper integrates the interaction time and social network into the graph neural network algorithm,and proposes a graph neural network recommendation algorithm(TS-LGCN)that integrates time weight and social trust.The main work of this paper is as follows:Firstly,a graph neural network algorithm is proposed,which integrates time weight to get the vector representation of users and projects based on historical interactive data.Construct a user-item interaction graph based on user historical interaction data.On this basis,the time weight value of interaction project is calculated by user interaction time.Then the time weight value and user score are integrated into the graph neural network algorithm,and the vector representation of users and items is obtained by high-order neighborhood aggregation.When getting the user's vector representation,the algorithm takes into account the user's interest in the project over time,and improves the accuracy of the algorithm.Subsequently,a graph neural network algorithm based on social trust is proposed to obtain a vector representation of users based on social networks.According to the user's social relationship,the user's social network graph is constructed,and the trust value between social friends is calculated by comparing the number of common interaction items and the threshold value between users.The social trust value is integrated into the graph neural network algorithm,and the user vector representation based on the user's social network graph is obtained by twolayer neighborhood aggregation.From the perspective of the user's social network,this algorithm can alleviate the user's cold-start problem of the model to a certain extent.Then,the neural network algorithm incorporating time weight and the graph neural network recommendation algorithm based on social trust are weighted and fused to obtain a recommendation algorithm integrating time weight and social trust.Finally,the yelp public data set is used to optimize the model parameters,and on this basis,the effectiveness of this algorithm is verified by comparing with five control algorithms.The experimental results show that the accuracy,recall and normalized loss cumulative gain of this model are higher than those of other control models.