Research And Practice Of Recommendation System Based On Graph Neural Networks

When it comes to solving the problem of information overload,recommender systems have become the primary solution,and its main task is to use various algorithms to model and analyze according to users’ historical interactive data,so as to predict users’ favorite projects and make personalized recommendations.Among them,collaborative filtering is one of the most important recommendation algorithms,but it has two major problems,namely cold start and data sparsity.Although the traditional deep learning algorithm can alleviate these two problems by incorporating auxiliary information,it is difficult to capture the structural information between the data.In this thesis,A recommendation system framework based on graph neural network is proposed,which is based on graph data,and improves the accuracy of recommendation by learning the relationship between nodes and adds some auxiliary modules to improve it.The main work accomplished in this thesis is as follows:(1)This thesis proposes the GNNFC(Graph Neural Network Based Feature Crosses).The algorithm can process the graph structure data that integrates the user’s attribute information and the attribute information of the project,use the graph convolutional network to learn the relationship between nodes,and finally predict the items that the user may like.(2)Considering that nodes can influence each other.In order to capture the high-order and nonlinear relationships between nodes,this thesis designs two new GNN aggregators for GNNFC,which enables the neighbor nodes of the target node aggregate messages in a node-crossing manner,so that the model can learn the highorder information between the neighbor nodes.(3)In order to learn the importance of the interaction between different nodes and reduce the interference of useless node crossover to the model,this thesis designs a graph attention mechanism in the process of message aggregation,which is used to learn the importance of node crossover.Further improve the recommendation performance of the model.(4)This thesis conducts extensive experiments on two representative datasets,comparing multiple baseline methods.Experiments have proved that GNNFC is superior to other algorithms in terms of accuracy and can be applied to different recommendation scenarios.(5)Based on the above research,this thesis designs and implements a recommendation prototype system based on graph neural network,and gives an application demonstration based on a movie scene.Combining the past recommendation algorithm and the new graph neural network algorithm,based on Spark and other Framework,designed each functional module,introduced and demonstrated its various functions.The successful use of the system verifies the feasibility and effectiveness of the model proposed in this paper.