Network Representation Learning Based On Improved Louvain Algorithm And Deep Autoencoder

Network data,a representation,is used to describe the relationship between objects.In the real world,various systems can be represented by network data,such as logistics network systems,biological information network systems,etc.However,these networks tend to have the characteristics of huge data and complex structure.Therefore,complex networks are often not fully acknowledged and understood by people.The study of network representation learning is one of the most important contents of complex network analysis.It helps people to acknowledge and understand the internal mechanisms of complex networks,to realize the application and development of complex networks,such as the prevention of infectious diseases in biological networks,public opinion monitoring in social media network,etc.Therefore,the study of network representation learning in complex networks is very important and of great significance.The node information in the network generally presents high-dimensional data characteristics,but it often cause difficulties in network analysis tasks.Network representation learning is mainly to transform the node information in the network into a low-dimensional vector representation,then to better analyze the network structure and the relationship between nodes.However,with the increasing size of the network,traditional learning methods cannot efficiently solve the problems of large-scale networks,so how to learn low-dimensional vector node representations in complex networks is very important.In order to solve all the problems,this paper introduces deep learning methods into the research of network representation learning,the purpose is to improve the network representation learning ability.Based on deep learning and network representation learning,the paper proposes two network representation learning algorithms,that is,Fast network folding method based on louvain algorithm and method a deep autoencoder network representation learning.The first method is mainly to solve the problem that the existing network representation learning methods are difficult to apply to large and complex networks;however,the second method is mainly to extract low-dimensional eigenvectors representation for each node in the network.The experimental results on different data sets prove that the methods proposed in this paper are more efficient and scalability than some classic network representation learning methods.In summary,the main research work of this article includes:(1)As the amount of data in daily life continues to increase,the scale of the network is becoming more and more complex,and the structure is larger,so there is an urgent need to design representation learning for complex networks.This paper proposes a fast network folding method based on the louvain algorithm.Utilizes the attributes of nodes in the network and aggregates large and complex networks into a super node through the community relationship between nodes,and a series of super nodes are reconstructed into a new net work after aggregation.The scale is much smaller than the original network.It not only preserves the local hierarchical nature of the network,but also preserves the overall nature of the network structure.For small-scale networks,the use of representation learning methods can learn lowdimensional feature representations,and it can have a certain effect on the current network representation learning based on large-scale networks.(2)In practical applications,networks usually contain a large number of nonlinear characteristics,and traditional network representation learning methods can be limited in real networks.In response to the above problems,This paper proposes a network representation learning method based on deep autoencoders.An effective method of network adjacency matrix transformation is used to describe the similarity of network nodes;the low-dimensional feature extraction is performed on the deep autoencoder to obtain an effective low-dimensional feature representation;the obtained low-dimensional feature representation used in downstream tasks,such as node classification,important node discovery,etc.In addition,this method can also be used as a method of embedding each layer of network nodes in a fast network folding method based on the louvain algorithm,and to achieve a more effective and accurate representation of nodes in a complex network.