Study On Recommendation Algorithm Of Graph Convolution Based On Knowledge Graph

In recent years,the development of graph-based convolutional neural network technology has enabled the node information in the knowledge graph to be trained to obtain a more semantic representation.At the same time,some work has shown that the knowledge graph can enhance the recommended items in the data set.The entity plays a certain role in optimizing the recommendation algorithm.To this end,with the recommendation algorithm based on the knowledge graph convolution neural network as the main research goal,research on the improvement of graph data,convolution calculation,and graph-based recommendation methods are carried out.This study uses extensive structural information of the knowledge graph to supplement user and project interaction data,and trains the graph convolution neural network on the user and project interaction graph after integrating the project attributes in the knowledge graph to obtain the representation of the project and the user.And recommend it.In the specific graph convolutional neural network node update process,that is,the target node collects the features of adjacent nodes to update,it proposes two training methods using asymmetric horizontal and vertical convolution kernels,and has obtained excellent recommendations in related directions result.The proposed methods include knowledge graph representation learning,graph convolutional neural network,attention layer and recommendation algorithm.Integrate the knowledge graph and user item interaction data,and use the existing representation learning method to train to obtain the vector representation of nodes and edges in the integrated graph.In the graph convolutional neural network layer,when training a node representation,the attention weight of this node and its first-order neighbor nodes is calculated according to the attention mechanism method,and then a certain number of adjacent nodes are sampled according to the weight,and horizontal and vertical scroll The product extracts features from these adjacent nodes to update the target node.In the final recommendation algorithm,the user and project information that have been trained are used to calculate the probability that the user will be interested in different projects at the next moment.The proposed algorithm uses knowledge graph as recommended auxiliary information,and innovates the process of graph convolution update node.Finally,through experiments,on two authoritative data sets,the recommendation results trained based on the horizontal and vertical convolutional graph neural network obtained good performance in the relevant direction.