Research On Quantification Of Network Structural Dissimilarities Based On Graph Embedding

In recent years,the research on complex networks presents a rapid growth trend,and how to quantitatively compare whether two networks are similar has also attracted more and more attention.There are many solutions to the network comparability problem,such as the comparison of topology properties based on the network,the comparison of graph classification methods based on machine learning,and so on.The comparison of topological properties of networks is only to compare some local characteristics between two networks,and the effect of the comparison is often unsatisfactory.The graph classification method based on machine learning is basically to give the type label of the network,and then train the network to classify the network.Although the classification method based on machine learning graph has a good effect,it has a long running time,and can only compare whether different types of networks are similar,and cannot accurately describe the differences between similar networks.Based on this,in view of the defects of the above methods,combined with the rapid development and wide application of graph embedding methods in recent years.This paper proposes a network difference quantization method based on graph embedding,which has strong generalization ability and can capture the global characteristics of the network.The main work of this paper is as follows:(1)In this paper,we propose a method for quantifying network structural differences based on graph embedding.The method works by representing each node in the network as a vector,and these vectors are further used to compare the similarity of the network.This method can capture the global characteristics of the network effectively,and the two terms of the network distance formula proposed in this paper compare the global similarity and the local similarity respectively,which makes up for the defect of the traditional method that only consider part of the network attributes and only capture part of the network information.Therefore,the model method in this paper can more accurately and reasonably capture the differences of topology structures among networks.(2)In order to verify the effectiveness of the proposed method,we conduct simulation on artificial network and real network respectively.For random networks,our model can effectively capture the phase transition points of random networks and compare the variation differences among small world networks.In simulation experiments based on real networks,we compare the differences between the node distribution obtained by the proposed method and the distance distribution of real networks,thus verifying the effectiveness of the proposed method in capturing global network features.Then we use the comparison of two different methods to verify the effect of our method in the real network.In one method,the differences between the real network and the network transformed by the null model are compared,while in the other method,the differences between the real network and the network transformed by random deleting and adding a certain proportion of edges are compared.These two methods can both verify the effectiveness of this algorithm model in the real networks.(3)We also compare the methods with those proposed in other papers.We compare the results of the three methods on four kinds of artificial networks.And we also experiment on artificial networks and real networks respectively.It is proved that our method achieves the best results in dense networks.(4)In order to verify whether the model can be extended to the quantitative comparison of networks of different sizes,this paper conducts experimental simulation on ER random networks with different node sizes.The results show that our method can also be used to compare networks of different sizes.Finally,we verify the influence of weight parameters on the model for WSL,ER,WSH and BA artificial networks.The experimental results show that the proposed model has good robustness for parameters.In general,the structural difference quantification method based on graph embedding proposed in this paper can not only compare the differences between artificial networks,but also compare the differences between real networks.It is not only applicable to compare networks of different network sizes,but also does not rely heavily on weight parameters.The model method in this paper provides a reference basis for the quantitative comparison of network in real life.