Research On Link Prediction Algorithm Considering Node Link Characteristics

With the continuous development of Internet and information technology in recent years,a variety of complex network systems appear.With the emergence of these complex networks,the link prediction index,which can be used as an algorithm to predict the change or completion of complex networks,has very important theoretical and practical application value,and has aroused extensive interest of researchers.The traditional link prediction indicators mainly include common neighbor based link prediction indicators,path based link prediction indicators and random walk based link prediction indicators,including CN index,AA index,RA index,Katz index,LRW index and so on.These classical prediction indexes can achieve the prediction function to a certain extent,but they usually have some problems,such as insufficient prior information mining contained in training data,large performance gap in different networks,poor stability and so on.In order to solve the above problems,the main work of this paper is as follows:This paper introduces and analyzes the background and significance of link prediction in complex networks and complex networks,and investigates and introduces the current research status of link prediction index.In order to provide theoretical support for the subsequent improvement and introduction,this paper briefly introduces the basic concepts,topological properties and several classic link prediction algorithms of complex networks.The defects of six traditional link prediction indexes,including CN index and so on,are analyzed,and a link tightness index called lplt index is proposed.This index not only considers the situation of common neighbors between link nodes,but also considers the closeness between common neighbors and link nodes,adding 2 On the other hand,an adaptive similarity transfer operation is introduced into the similarity index to solve the similarity estimation between the link nodes without common neighbors.Through the simulation of AUC index performance under different network topology and different training data proportion,it can be found that lplt algorithm has better performance than six traditional link prediction indexes under different network and data proportion,which shows that lplt algorithm can effectively adapt to different network and training data sample number,and has better link prediction performance and reliability.Aiming at the instability of the traditional Katz prediction index,an improved Katz prediction index is proposed,which can more effectively predict the reliability of the index;secondly,aiming at the situation that the traditional prediction index can not be adjusted adaptively according to the network,an adaptive mismatching method is proposed by learning the trained network AKL index of the same network characteristics.This paper also analyzes the AUC performance and precision performance of AKL index in different network topologies and different proportion of training data through simulation.Both the overall and local AKL prediction indexes show better prediction performance.At the same time,it effectively solves the problem of unstable performance of traditional algorithms in different networks,which shows the effectiveness and reliability of the prediction index It provides a new effective and reliable solution for link prediction in complex networks.