Link Prediction Algorithm Based On Higher-order Features Of Nodes

Since Watts and Strogtz proposed the small-world network model in 1998 and the scale-free network model proposed by Barabási and Albert in 1999,which led to the explosive development of complex network science.The researchers have built complex network models for various complex systems in reality,such as traffic networks,ecological networks and social networks.In order to solve pratical matters,such as traffic jams and ecosystem protection,scholars began to study the behavior and characteristics of various complex network systems,such as the evolution mechanism,connectivity and invulnerability of complex networks,and found that these behaviors and characteristics are inseparable from the behavior of individuals in complex systems,and it is more inseparable from the relationship between the individual and the individual.Link prediction provides an effective prediction mechanism for mining the association between individuals.Therefore,in order to mine the behavior and characteristics of complex networks more efficiently,the researchers propose many different types of link prediction algorithms.By comparing these link prediction algorithms,it is found that they rarely consider the high-order similarity relationship between nodes,and the link prediction algorithm based on the low-order similarity relationship of nodes has poor link prediction performance.Based on this,three link prediction algorithms based on the high-order characteristics of nodes are studied,which improve the performance of link prediction in different aspects.(1)A link prediction algorithm based on high order proximity approximation is proposed.This method combines the high-order network representation learning algorithm with link prediction,and considers the high-order similarity relationship between nodes.At the same time,through the experimental simulation on four real datasets,the results show that the link prediction algorithm based on high order proximity approximation has better prediction performance.(2)A link prediction algorithm based on the gravitational field of complex network is proposed.This method considers the high-order characteristics between the target node and the other nodes in the network in physics terms,abstracts the nodes in the complex network into the mass points in the gravitational field,and takes the nodes' degree values as the method to evaluate the importance of the nodes,and then constructs the gravitational field model of the complex network,finally applies it to the link prediction.After experimental simulation,the prediction algorithm based on the gravitational field of complex network shows better prediction performance.(3)A link prediction algorithm based on the gravitational field of complex network and node contraction is proposed.On the basis of the link prediction algorithm based on the gravitational field of complex network,this method takes node contraction as a way to measure the importance of nodes,and puts forward an improved gravitational field model of complex network and applies it to link prediction.Through experimental simulation,on the basis of the experimental results of the link prediction algorithm based on the gravitational field of complex network,the results show that the link prediction algorithm based on the gravitational field of complex network and node contraction has also been improved to a certain extent.