Research On Representation Learning Of Heterogeneous Networks Based On Structure Walk

Heterogeneous network representation learning is the process of learning lowdimensional dense and independent vector representations in a network.Most researches use random walks to traverse nodes.However,it is not suitable for large and complex networks and can not ensure the optimal path information.Therefore,in order to solve the above problems,this paper proposes a heterogeneous network representation model based on structure walk.The specific research contents are as follows:Firstly,in order to solve the problem that the random walk method is difficult to express the complete information of nodes and the optimal solution of vectors,propose a method based on structure walk.A unique vector representation of the node is generated by using the relationship distance between the node and other nodes,so as to completely represent the location information and semantic information of the node.Secondly,for the problem that the vector representation of nodes is long and sparse,an autoencoder model suitable for representation vector compression is proposed.The lowdimensional representation vector obtained after compression is used as the representation vector of nodes,and is verified by experiments through classification and clustering tasks.Thirdly,in order to improve the compression performance of the above autoencoder compression model,add the residual network and convolution attention mechanism.Using the improved autoencoder to do classification and clustering experiments again,and the results are compared with the above results.Finally,in order to further extract important informative features,drawing on the idea of generative adversarial network,add a discriminator structure.The features of the vector are extracted by using the autoencoder with the discriminator structure,and the comparison and analysis are carried out through node classification and clustering experiments.