Research On Recommendation Algorithm Based On Graph Neural Network And Attention

With the popularization and development of the Internet and mobile terminals,network information grows explosively.How to accurately and quickly help users obtain interesting information has become the main research goal of recommendation systems.Based on different user and item Representation Learning methods,this paper proposes two recommendation algorithms based on graph neural network: Graph Neural Networks on Collaborative Filtering Recommendation via Attention(GCFA)and Incorporate Sessions into GNN-based Recommendation algorithm,ISGNN)both algorithms use graph neural network to learn embedding.GCFA proposed a rating prediction recommendation model based on graph neural network.First of all,this article provides a way to jointly capture the interactions and opinions in the user-item graph.Secondly,the GCFA takes into account the different strength of user-item interactions,suggesting that various interactions contribute differently to user preferences(or item characteristics),using attention mechanisms to achieve this goal.Next,the embedding propagation layer is used to encode the collaborative signal.Finally,the results of Ciao and Epinions show that GCFA is better than other recommendation algorithms.ISGNN incorporative the idea of session into the recommendation algorithm based on graph neural networks.Firstly,for the user-item bipartite graph,the one-mode projection is used to obtain a compressed user-user relationship graph.By comparing similar users of the user,a coarse-grained representation of the user is obtained,which is the potential intention at the user level.A fine-grained embedding representation of a session is obtained by combining the long-term and short-term interests of the user session.Furthermore,the combination of coarse-grained user intentions and fine-grained session representations can provide more accurate user embedding representations,so as to better provide personalized recommendations.It is verified by the public dataset that this algorithm has better recommendation effect than other classical recommendation algorithms.This paper first introduces the background and significance of the topic,and analyzes the commonly used recommendation algorithms at present.Then the principle and modeling process of two recommendation algorithms based on graph neural network are introduced in detail,and the validity of the proposed algorithm is verified on the public datasets.Finally,summarize the author's work during the paper and prospected the future research direction.