Design And Implement Of Collaborative Filtering Recommendation System Based On Graph Convolution Neural Network And Attention Mechanisms

With the rapid development of the Internet,information redundancy is becoming more and more serious.How to extract useful information from a large number of redundant information has become a hot research topic.Different selection schemes are provided for users according to their historical behavior data,which often ignores the personalized characteristics of users,thus making the selection scheme difficult to reflect the real intention of users.Therefore,the traditional recommendation model is improved to meet the personalized recommendation requirements by combining user profiles,context information and other methods.This thesis mainly describes the technical principles involved in the research of graph convolutional attentional neural collaborative filtering recommendation system,and will focus on the principle and process of combining graph convolutional neural network and attentional mechanism,and also discusses the number of iterations and iteration methods for the full graph information to improve the propulsion performance,solve the data sparsity,add user portraits,and realize the graph convolutional recommendation system based on the traditional recommendation algorithm.attentional neural collaborative filtering recommendation system.The main work of this thesis is as follows:(1)In this thesis,based on the inability of traditional recommendation algorithms to effectively solve problems related to data sparsity and recommendation accuracy,this thesis proposes an adaptive graph convolution attention neural collaborative filtering recommendation method(ANGCACF).ANGCACF algorithm firstly obtains user and item interaction graphs,and adaptively aggregates user and item feature information by graph convolutional neural network;secondly,adds adaptive expansion data to user and item feature information to solve the problem of Secondly,it adds adaptive expansion data to the user and item feature information to solve the data sparsity,and reassigns weights to the user and item feature information and the added adaptive expansion data by using the attention mechanism;finally,the obtained user and item feature representations are used to derive the final recommendation results using the algorithmic framework of collaborative filtering based on matrix decomposition.(2)Based on the high time complexity and space complexity of the ANGCACF recommendation algorithm,this thesis proposes an attention based Graph Convolution Attention Collaborative Filtering(GCACF).The GCACF algorithm first obtains the relevant interactive information of users and projects,and converts it into corresponding feature vectors;Secondly,the eigenvectors are aggregated with localized information using the propagation mode of graph convolution neural network(GCN),and the aggregated weight coefficients are redistributed using the attention mechanism;Finally,the aggregated eigenvectors are used to optimize the relevant parameters using the BPR loss function and the final recommendation results are obtained.The essential difference between GCACF algorithm and ANGCACF algorithm is that ANGCACF’s iterative method based on adaptive GCN and attention mechanism can iterate all the information in the graph into new feature information,while GCACF algorithm uses GCN and attention mechanism to ignore some information to ensure fast response and offline training time of the model.The proposed algorithm and benchmark algorithm are experimentally analyzed and summarized on three public datasets.The experimental results show that the proposed algorithm is superior to the baseline method in five indicators: recommendation accuracy,recall,MRR,hit rate and NDCG.The algorithm can process non Euclidean structural data,improve recommendation accuracy and data sparsity,and analyze and compare the differences and differences between the proposed two recommendation algorithms.(3)This thesis uses Spark,Spring boot,Redis,Flink,Tensorflow and other related technologies to implement a graph based convolutional attentional nerve collaborative filtering recommendation system,including demand analysis,related technologies,system design and system display.Through this system,you can directly view the recommendation results of GCACF algorithm and ANGCACF algorithm.