Research And Application Of Vector Space Embedding Technology Of Attribute Graph

With the high-speed development of the Internet today,massive network graph structure data has emerged,and these graph structure data often hide a wealth of valuable information,which has high potential commercial value and important significance.However,because the graph structure has the characteristics of abstraction and complexity,it is necessary to first convert the network graph into a vector space,and then use the method of machine learning to perform subsequent processing on it,in order to effectively dig out valuable information.How to accurately embed network graphs into low-dimensional vector spaces has become a hot field and topic in computer science research.At present,most typical network graph embedding algorithms only focus on the structural information of network graph nodes,and ignore the influence of node attribute factors on the results of network graph embedding.The more influencing factors considered,the more complete the original information of the network diagram retained in the results after Embedding,which is crucial to the network diagram embedding algorithm.Therefore,the attribute graph embedding algorithm AS2Vec(Attribute and Structure to Vector),which comprehensively considers the structure and attribute characteristics of the network graph nodes,came into being.The algorithm uses matrix fusion to fuse the structure correlation matrix and the attribute correlation matrix of the graph nodes.Then a network graph containing structure and attribute context information is obtained,so that taking into account the structure and attribute factors of the network graph node at the same time,the reconstructed network graph is used to learn the vector representation of the network graph node in the low-dimensional space.The network graph embedding algorithm is applied in two real scenarios of node classification and link prediction.According to the accuracy of classification and prediction on related data sets,the embedding effects of AS2 Vec,Deep Walk,LINE and other graph embedding algorithms are compared and analyzed.Compared with other typical graph embedding algorithms,the accuracy of AS2Vec's classification and prediction has been improved,which proves the superiority of the algorithm,and also proves the operation of using node attributes as additional information in the process of graph embedding is feasible.