Link Prediction Based On The Structure Of Complex Networks

With the rapid development of the Internet,the structure of real-world networks has become more complex.In many studies on complex networks,link prediction research has become a key tool for mining network structures and evolution mechanisms,and is widely used in many disciplines.As a long-standing practical scientific problem,link prediction research helps us to understand the mechanism of complex network evolution in theory and has important applications in identifying lost and false links.In the past two decades,network science has become a new architecture for understanding the structure of many real-world network systems.Link prediction,as an important branch of network architecture analysis,has attracted many resear-chers in various fields.Scholars in different fields have applied various discipline technologies to link prediction research,hoping to enhance the overall prediction accuracy of the prediction model.This paper is devoted to further mining the characteristics of network structure,and proposes thr-ee link prediction models with better prediction performance.The main innovations and works in this paper are as follows:(1)A link prediction model based on non-equilibrium cooperative effect is proposed.Most traditional network-based link prediction models do not consider the heterogeneity of endpoint relationships,making these models ineffective for various types of networks.This paper explores the non-equilibrium cooperative relations existing in the network and analyzes the influence of this effect on the formation of potential joints between nodes.We propose a link prediction based on non-equilibrium cooperative effect by introducing the degree heterogeneity index H to penalize the large-degree nodes in different networks.The experimental results on twelve real-world network datasets show that the link prediction model based on non-equilibrium cooperative effect can be adapted to different heterogeneous networks and effectively improve the comprehensive performance of link prediction.(2)A link prediction model based on node centrality is proposed.Most existing link prediction models use node degree as the quantitative indicators of node influence,but the network information used by node degree is limited,and the measurement of node influence is not accurate enough,thus affecting the prediction accuracy of these link prediction models.This paper introduces the closeness centrality and betweenness centrality to quantify the importance of nodes and proposes a link prediction model based on node centrality.The proposed model tradeof-fs the quantization accuracy and computational complexity.The experi,mental results on six real network datasets show that the link prediction model based on node centrality effectively impro,ves the prediction accuracy.(3)This thesis proposes a link prediction model based on path transitivity.Most of the existing path-based link prediction models simply consider the number of paths between two endpoints,and rarely analyze the path structure.The isometric path of different structures has a negligible difference in the ability to transmit similarities between two edfnodes.Based on this,we propose a link prediction model based on path transitivity.The model considers that the short path with small influence transition nodes has stronger endpoint similarity transfer capability.What's more,according to different network structures,the heterogeneity of the path is considered.After repeated verification on six real network datasets,the experimental results show that the proposed model effectively improves the accuracy of prediction.