Study On Link Prediction Based On Network Embedding And Transfer Similarity Methods

With the rapid development of information technology and the emergence of complex systems,network science has also been developing rapidly as an emerging discipline.Link prediction is an important topic in this field.Based on existing network topology and other information,it can find edges that are not yet known and predict edges that do not exist but may emerge in the future.In practice,link prediction can be used to recommend friends in social networks and to find unknown links in gene regulation networks.Theoretically,it helps to reveal the internal structural characteristics of the network and understand the evolution mechanisms of complex systems.The current link prediction algorithm mainly predicts edges according to node attributes or network topology.In practice,the former has the problems that node attribute information is difficult to obtain and even they are obtained,their quality is not guaranteed.Though the latter only needs to know the topology of the network,its accuracy is usually low.The research in this paper belongs to the latter,that is,we make predictions only based on network topology.This paper first proposes an Adjacency Embedding(AE)algorithm based on High Order Proximity preserved Embedding(HOPE)algorithm.It characterizes the nodes in the network with low-dimensional,real-valued,dense vectors.By calculating the distances between these vectors in the low-dimensional space,the similarity of the nodes is inferred,and the possibility of edges existing between them can be predicted.Through experimental analysis on 10 real networks,we can conclude that compared with HOPE algorithm,AE algorithm can reduce time comsuption while maintaining the prediction accuracy.Secondly,based on the idea that similarity should be transferable,we propose a method called TSBAE(Transferring Similarity Based on Adjacency Embedding)based on the AE algorithm,which increases the accuracy of predicting similarity between nodes that are far apart from each other.The results show that the TSBAE algorithm has higher prediction accuracy than the baseline algorithms and AE algorithm on the undirected networks and directed networks.