Research On Attributed Community Search Method With Hidden Relationship

With the rapid growth of the number of users of online social networks,large-scale graphs are characterized by high dimensions and data sparsity,which makes mining valuable information from these graphs a challenging problem.Network embedding maps nodes in the graph to low-dimensional space and uses the similarity between low-dimensional vectors to represent the relationship between nodes,which makes it possible to analyze and mine largescale graphs.The low-dimensional vectors obtained from network embedding can be applied to downstream tasks such as node classification,link prediction and network visualization.In addition,the real network is often a kind of attributed network,because the nodes are often rich in attribute characteristics.Therefore,it is necessary to study the representation learning method of attribute network.The community search problem only concerns the community where the query node is located,which greatly simplifies the scale of the problem,making graph search under big data possible.Property community search is to find a community with specific properties where the query node is located.The existing community search algorithm only studies the community structure of dense subgraphs,but because of the incomplete graph data itself,the search results differ greatly from the real community.At the same time,the existing community search algorithm with the query node as the center of the community may cause the community migration problem.In view of the above problems,this paper studies the network embedding model and the attributed community search based on the model.The specific work mainly includes:Firstly,an attributed network embedding model is proposed,which can extract the structure and attribute characteristics of the network,and combine the structure and attribute characteristics nonlinearly to obtain the low-dimensional representation matrix of the network.The related concepts and descriptions of the attributed network embedding problem are given,the structure of the model,the transfer process of the loss function and data between the layers of the model are described,and the training steps of the model are given.Finally,the experiment was carried out on the data sets of Blog Catalog3,Facebook and DBLP,and the learned presentation vector was applied to the three tasks of graph reconstruction,multi-label classification and link prediction,which were compared and analyzed with the existing models.The experimental results verify that the model presented in this paper can better represent the learning effect in the attributed graph.Secondly,based on the above network representation model,an attribute community search method with hidden relationships is proposed.Based on the description of attribute community search problem,the basic idea and formal description of attribute community search method are described in detail.By adding community offset correction algorithm and hidden relationship discovery algorithm,this method improves the existing community scoring method and solves the problem of community offset and incomplete graph data.In the experimental part,four comparative experiments were carried out.Experimental results verify the correctness and feasibility of the method proposed in this paper.Through comparative analysis of experimental results,the community search method proposed in this paper performs better on F1 values than other community search algorithms,which verifies the feasibility and correctness of the proposed method.