Representational Learning And Application For Structures Equivalence On Complex Network

The rapid development of Internet technology has made complex network research become a hot spot today.In recent years,network embedding has become more and more popular for network analysis of complex networks.However,today's representational learning methods lack attention to the structure equivalence of the network,and use the structure equivalence of the network for risk calculation has proved to be an effective method.Therefore,this article proposes a risk learning method that uses both the global and local structure equivalence representation learning methods.The specific work is as follows:Firstly,a representation learning algorithm(DMER)for the mutual enhancement network structure equivalence is proposed.The method firstly extracts the feature of the structural feature information of the network,then uses the graph convolution network and the auto-encoder to model the structure equivalence of network from the global and local aspects,and then enhances each other with a parameter shared decoder.The model can retain both global and local structural information of the network.The evaluation of the effects on public data demonstrates the practicality of our approach.Secondly,a representational learning algorithm(DCER)is proposed for the equivalence of cross-attention network structure.The model is modeled using the graph attention model according to the influence of the node neighbors on the structure of the network.And using the cross-attention mechanism to make the variational automatic encoder and the graph attention mechanism model better mutually increase.In turn,the model has good scalability and can be used in large-scale networks.Finally,using mobile phone signaling data to conduct empirical analysis on the detection of fraudsters and pyramid schemes from four aspects.The user's structural equivalence representation from the four networks is learned,and the user's potential risk portrait is drawn by comparing with known risk users.Use the potential user risk image obtained from the four data to draw a potential risk picture on the map based on its latitude and longitude.The results of the above work show that this topic can indeed divide different roles according to the low-dimensional representation of its nodes according to the structural equivalence of the network.That is to say,in the mobile phone data,the model of this paper can identify the potential risk areas according to the network structure characteristics of the mobile phone network,and then provide important information support for the community security for decision makers.