Research And Application Of Locally Enhanced Attribute Network Embedding Via Deep Auto-encoder

As a carrier of information,social network not only contains rich information in the link relationship of nodes,but also has diverse information itself.Therefore,data mining of social networks is of great significance.With the popularization of Internet,the number of nodes of social network increases gradually,and the network structure becomes more and more complex,which brings unprecedented challenges to the existing network mining technology.How to map sparse and high-dimensional network data to dense and low-dimensional feature vectors has become the key to solve the problem.Therefore,network representation learning model has become a research hotspot for a long time.Network representation learning refers to the use of some method to represent sparse,high-dimensional original input data as dense,low-dimensional vectors,while retaining network information and discovering the inherent law of data.With the emergence of deep learning technology,there are some presentation learning models based on depth models.SDNE(Structural Deep Network Embedding)model is one of them.The SDNE model uses first-order and second-order nearest neighbor relations to maintain the network structure,and uses deep self-coding to learn the low-dimensional feature vectors of the network structure information.However,it only uses the structural information of the network,ignores the attribute information of the network nodes themselves,and fails to make full use of the paired constraint information in the network.Based on the SDNE model,the network attribute information was incorporated,the feature representation of network structure and attribute information was learned by deep auto-encoder,and the local information of the network was fully utilized to generate LEANE(Locally Enhanced Attribute Network Embedding Via Deep auto-encoder)model.This paper focuses on SDNE embedding model,and the main research contents are as follows:1.LEANE model study.On the basis of SDNE model,adding multiple depth selfencoders,learning low dimension feature representation of fusion network structure and attribute information;At the same time,we use Laplace feature mapping to realize local enhancement;Finally,LEANE model is formed.This model is used for network feature representation learning,and the learned low-dimensional feature vectors are classified and clustered.Compared with 7 similar models on 5 real data sets,the results are better than those of the same model.2.Model application.The data of CSDN social network is preprocessed,and LEANE model is used to learn the low-dimensional feature representation of CSDN social relations and attribute information.K-means,GMM(Gaussian mixture model),Birch(balanced iterative reduction and clustering using hierarchies)are used to complete the user clustering task,and the three clustering results are integrated to analyze the user Long type,and for each cluster of users to develop targeted marketing measures and services,to achieve refined operation.