Network Alignmet Based On Graph Neural Network

With the continuous development of the Internet,more and more social network users,people are increasingly keen to participate in a variety of social network activities.Therefore,how to find out the same users from multiple social networks is a very valuable task,which is also called network alignment.This task is of great value in cross network recommendation,sensitive crowd positioning and other work.Currently,the solution of network alignment is generally divided into two steps: first,social network embedding: because social network is a very complex structure,it is very difficult to perform calculation directly on social network.Therefore,the network is usually embedded into a low dimensional space,and a low dimensional vector is used to represent the nodes in the network.Second,Anchor link matching: after obtaining the embedded representation of the network,the common idea is to use the known anchor nodes to learn the matching function between two embedded representation vectors.Using this matching function,we can calculate the nodes in the source network that may be the same as each other in the target network.But at present,network alignment generally has the following problems: in the process of embedding,how to retain the network structure in the low dimensional space has become a very important problem.In the process of anchor node matching,the main problem is that the number of anchor nodes is usually small.How to efficiently use the known anchor nodes for matching learning is a very important problem.In this paper,aiming at the above problems,the following scheme is adopted: the graph variational self encoder is used to embed the network,and a better embedding result can be obtained by adjusting different parameters.At the same time,in the process of anchor node matching,a multi-layer perceptron architecture is adopted,which makes full use of anchor node to update the matching function by using multi-layer perceptron and its dual mechanism.Finally,an experiment is carried out on a real data set.The experimental results show that the alignment method proposed in this paper achieves a good result,and is better than some common methods in accuracy.At the same time,through the comparison of different parameters,some factors which have great influence on network embedding are obtained.