Multi-layer Network Representation Learning Based On Deep Walk

Network theory is an important tool for describing and analyzing complex systems in social,biological,physics,information,and engineering sciences.However,most of the existing theories are single,static descriptions of complex networks,which cannot describe the characteristics of the diverse interaction modes of complex networks.Multi-layer networks have been proposed to characterize complex networks.Therefore,in recent years,the multi-layer networks proposed internationally have become one of the important research directions in the field of complex networks.However,the data size of multi-layer networks is very large,and directly analyzing it will cause excessive calculation costs,low efficiency,and difficult to observe hidden patterns.Recently,network representation learning has been proposed to solve this problem.Therefore,learning the representation of multi-layer networks is particularly important for the task of network analysis.Multi-layer network representation learning aims to preserve the unique structure and characteristics of multi-layer networks.The Deepwalk method has shown good performance in single-layer network representation learning.Therefore,this paper is based on the Deepwalk method for multi-layer network representation learning tasks.This paper mainly considers the representation of learning nodes from the relationship between the layers and nodes of the multilayer network and the community structure.1.First,this paper proposes a multi-layer network representation learning method that can maintain the unique network structure information of each layer of the multi-layer network and retain the feature information shared by the nodes in the multi-layer network.This method introduces the concept of layer vector and node shared feature vector to capture the structural features of each layer and shared features of the same node between layers,and introduces the similarity between layers to further refine the node characterization and layer characterization.This paper also compares this method with other methods on real datasets,and the experimental results show that this method shows better or equivalent performance.2.Second,this paper also proposes a multi-layer network representation learning method that incorporates community structure features.This method introduces a deep walk algorithm in the form of matrix decomposition to model the topological relationships within layers in a multi-layer network.At the same time,the relationship between cross-layer community structure and node representation in a multilayer network is considered.Finally,all the above information is considered to form a unified framework,and the multilayer network representation is obtained through optimization learning.And the comparison experiment on the real data set verifies the effectiveness of the method.