Abstract: As a research hotspot in complex network analysis,link prediction has a wide range of applications in many fields,and hence has captured much attention of researchers.Similarity-based methods,which compute the similarity scores between unconnected node pairs based on the known network structures and estimate their connection likelihood according to the similarity scores,are commonly used.In general,different kinds of networks have diverse structural characteristics,and hence the correlation of characteristics between nodes has an important influence on the formation of links.To enhance the performance of link prediction,this paper defined the connecting pattern of a node,and proposed a new link prediction model based on the Correlation of Nodes’ Connecting Patterns (CNCP).By combining CNCP with a similarity-based method,this model can take both similarity and correlation between nodes into account.In this paper,four CNCP-based methods,i.e.,CNCP-CN,CNCP-RA,CNCP-AA and CNCP-PA,are derived from the model,in which similarity indexes are CN(Common Neighbors),RA(Resource Allocation),AA(Adamic-Adar) and PA(Preferential Attachment),respectively.The experimental results on six networks show that the proposed methods are superior to the compared ones under the criteria of AUC and Precision.

Key words: Complex networks, Correlation of nodes’ connecting patterns, Link prediction, Similarity index