The Network Embedding Method With Direct Edge Content Integration

Nowadays,network representation learning as a critical tool for numerous network analysis tasks has attracted more and more attentions to many scholars.The purpose of network representation learning aims at learning a low-dimensional,dense,continuous,and reasoning attribute vector for each node in the network.The vector can not only measure the spatial relationship between nodes in the real network,but also deeply reveal the potential features between network nodes.In addition,the vector can be applied to other network analysis tasks as original features.At present,the main network representation learning strategy generally takes the network topology or node attribute content as the focus of the methods.The method based on network topology emphasizes the relationship between network original structure,such as first-order proximity,second-order proximity,high-order proximity,etc.The methods of capturing proximity are similar,mainly including random walk,adjacency matrix or sampling.For the attribute content of the node,the main methods rely on the tools of dimensionality reduction methods,such as natural language processing,deep learning,principal component analysis and so on.They encode the attribute content of a node into a semantic representation of the node,and the semantic representations are rich in semantic relationships.They generate a final node representation combined this semantic representation with the network representation generated by the network topology.As a part of the network content,the edge content has an important value for the network representation learning of nodes.However,it is still ignored by many main network representation methods.The edge content has strong similarities with the node content,so that it is easy to be mistaken as the node content to be processed.So the attribute content information of the edge cannot be fully utilized.In the main methods,there is an urgent need for a method of node representation learning method with the direct edge content integration.In order to achieve above objectives,this paper proposes a method of directly using the edge content information to enhance network representation learning.We expound a series of discussions on the correctness,effectiveness and complexity of the proposed model.In addition,this paper also designs some relevant experiments to verify the effectiveness of the new model on the real datasets.Considering the shortcomings of the proposed model,we also propose an effective solution to expand and supplement the model,which is used to extract the more deeper semantic information of edge content.