The Link Prediction Algorithm Research Based On Topology Of Networks

Link prediction is one of the most important researching directions in complex networks,it aims at predicting the existence of a link between two entities,which is based on attributes of the objects and other observed links.Link prediction has important applications in many fields,including recommendation systems,criminal networks,community detection,business decisions and so on.At present,the algorithms which are based on similarity are one of the mainstream research directions.In these algorithms,the higher the similarity between nodes in networks is,the more they tend to link.The algorithms which based on similarity have the advantages of low complexity and high prediction accuracy,and are suited for the link prediction in large scale networks.These algorithms can be divided into the attribute information of nodes based and the topological structure information of networks based.Due to the attribute information of nodes in networks is not easy to obtain,so the algorithms based on the topological structure information become the research hotpot.Most of the traditional algorithms which based on the topological structure information just focus on the number of neighbors or the degree of common neighbors of predicted nodes,and they not treat the local structure information of predicted nodes as a whole,which contains more useful information.In this paper,we analysis the existing algorithms,and use the theory of the strong community structure of complex networks to present Small Community index(SC).In SCs which considers the connections between the small community are more important,the predicted nodes and their common neighbors are considered as a small community.Then,based on SC and the results of real experiments,we make sure the longest length which maintaining the accuracy and keeping lower time complexity in small communities.In recent years,with the deepening research on complex network,scholars found that the directed and weighted networks can describe the real networks more accurate,and lots of researches have confirmed the rationality of the directed and weighted networks.Therefore,to the research of link prediction,we shouldn't just consider the existence of links in networks,but the directions and the weight of links.We should set up different networks by the properties of links in real networks.In this way,we can increase the accurate of link prediction and extend the theories and applications by folly using the information in networks.Thus,in this paper,we make Local Path Index and SC extend to the undirected weighted networks,and they can fully use the weight of links.Then we make Common Neighbors Index,Resource Allocation Index,Local Path Index and SC extend to the directed unweighted networks and directed weighted networks,we test these similarity indices in real networks with different properties,and the results of the experiments show that the extended indices can achieve good outcomes of prediction.