Research And Application Of Signed Network Embedding

The rapid growth of the Internet and social media has led to the rapid development of complex network analysis.In recent years,an effective method of network analysis:network representation learning has received extensive attention.Network representation learning aims to learn the low-dimensional vector representation of nodes.However,the traditional network representation learning method cannot be directly applied to the signed network.At present,the representation learning method proposed for the signed network only considers the structural balance theory of the signed network.However,for directed signed networks that are widely used in the real world,the theory of Status is often more suitable than the theory of structural balance.Therefore,this paper uses the Status theory to learn the representation of the signed network.The specific work is as follows:Firstly,a signed network representation learning method(SSNE)based on Status theory is proposed.This method uses the Status theory to learn the representation of the directed signed network for the first time,then designs an effective negative sampling method for the signed network and an enhanced attention mechanism for the node representation.Finally,the obtained node representation is used in the downstream machine learning task.Secondly,a representation learning framework(GATSNE)based on attention mechanism and graph neural network is proposed.In order to maintain the neighbor structure of the signed network and the social theory of Status,the framework uses a graph neural networks based on attention mechanisms to learn the highly nonlinear representation of nodes and designs loss function to maintain Status theory,and the sign prediction is also used to evaluate the effectiveness of the node representation.Finally,the model is applied to the field of alarm behavior analysis in urban areas.The signed network is constructed by using 110 alarm data and mobile communication dataset,and the related metric of alarm behavior analysis is defined from the perspective of network representation learning.Then the alarm behavior is analyzed in various ways.The experimental results illustrate that the network representation method proposed in this paper is effective in the analysis of alarm behavior.In this paper,a network representation learning method based on Status theory is proposed and a more general representation learning framework is designed by using graph neural networks.The obtained node representation is better than existing methods in tasks such as link sign prediction,so it has important theoretical value in the fields of network representation learning and signed network analysis,and the effect in the field of alarm behavior analysis reflects the practical value of this model.