Research On Link Prediction Method Based On Cluster Structure And Node Importance

With the continuous development of information technology such as the Internet and5G,more and more complex systems have emerged around us.Any complex system can abstract a complex network composed of interacting entities.Complex networks describe complex systems through nodes and their connected edges,and are important tools for modeling and analyzing complex systems.Link prediction is one of the important research branches in complex network research,which can predict the possibility of future connections between two nodes in the network,and can also identify false or lost connections in complex networks.In link prediction,most methods only consider low-order network structures and cannot fully mine network topology information.At the same time,the existing directed network link prediction methods have a relatively single weighting method.When calculating the similarity between nodes,they only focus on the common information between nodes,such as common neighbors,node degrees,paths,etc.,ignoring the contribution of the node’s own importance to the formation of new links.The main research content of this article is as follows:Firstly,this article focuses on the problem of high-order structural features without considering nodes.A link prediction algorithm based on clustering structure and common neighbor penalty（CCNP）is proposed.This algorithm determines the similarity score based on the higher-order structure of the network and the common neighbor penalty.Firstly,the common neighbor penalty factor is determined based on the network structure to improve the universality of the algorithm.Then,high-order clustering coefficients of nodes are introduced to distinguish the differences between nodes in the common neighbor set,quantifying the contribution of each neighbor node to link formation,and further improving the accuracy of the algorithm.Secondly,this article addresses the issue of not considering the contribution of node importance to the formation of new links.A weighted link prediction algorithm based on node importance for directed networks（WPDN）is proposed.This algorithm comprehensively utilizes the common neighbor information of nodes and the importance of nodes themselves.Firstly,calculate the first-order h_in index and h_out index of nodes in the directed network,and then evaluate the importance of nodes based on their degree.Then,the importance of nodes is integrated into the adjacency matrix of the network by means of reciprocal link weighting,and Bifan index is improved and applied to the reciprocal link weighting network to calculate node similarity.In addition,the WPDN idea has been applied to the classical directed network weighting metrics DCN,DAA,and DRA.Finally,the link prediction algorithm based on clustering structure and common neighbor penalty and the weighted link prediction algorithm based on node importance for directed networks are each experimentally validated on different types of real data sets.The two algorithms are also compared and analyzed with classical link prediction methods to verify the effectiveness of the algorithms proposed in this paper.