A Research On Community Analysis Method Based On Network Embedding

The complex network with content information of nodes(attribute network)is widely used in daily life,for example,social networks,citation networks,and protein networks.The community analysis task in the attribute network can solve the friend recommendation,content recommendation and other real problems.However,in real life,the information sources of nodes' attributes and structural features are often quite different.Generally,the attribute feature dimension and the number of nodes are also large.It is very difficult to use the classical machine learning method represented by K-means to solve the community analysis problem of attribute networks.Network representation learning(network embedding)aims to represent nodes in the network as low dimensional real value vectors and to some extent retains the intuitive and potential information in the network.Therefore,it is of high research value to combine network embedding method to solve the community analysis problem in attribute networks.At present,the mainstream network embedding methods are mainly implemented based on neural network model,and the representation learning of the node's attributes and structural characteristics is not sufficient,so when the output of network embedding method are applied to community analysis,the accuracy and recall rate are low.To solve these problems,taking attribute network as the research object,a network embedding algorithm based on attribute and structure(Network Representation Learning Algorithm Based on Attribute and Structure,NRLAS)is proposed.First,NRLAS algorithm designs calculation constraint of node correlation degree based on attribute similarity and structural similarity of nodes and the node correlation degree is taken as the precondition of random walk probability to make the sampling effect of random walk on the whole network more sufficient.Secondly,NRLAS algorithm proposed a neighborhood extension method,so that the node neighborhood is not limited to the network structure,and the neighbors in the attribute space can be used as a valuable reference.When expanding the neighborhood,the neighbors in the network structure and the neighbors in the attribute space can be combined to make the network embedding to learn the feature of community structure in the attribute space.Finally,the learning model of NRLAS algorithm is built based on graph convolutional neural network,and the performance of the algorithm is improved by integrating the attribute features of neighbor nodes and the current node attribute features.Extensive experiments on multiple real-world datasets(such as Cora,Cite Seer and Pumbed)with varying sizes and characteristics demonstrate that the accuracy and recall rate of NRLAS algorithm are significantly improved compared to other classical methods on node classification and node clustering tasks.NRLAS algorithm provides a new idea for solving community analysis,which has good theoretical and practical properties.