Research Of Link Prediction Algorithms Based On Similarity In Social Networks

In the study of social networks,link prediction has become a very important research direction.Using network topology,link prediction can predict the possibility of the connection between two nodes unlinked at present.In social networks,link prediction can not only predict missing links,but also predict future links,which makes it valuable both in theory and practice.In the many link prediction algorithms,the link prediction algorithms based on similarity have become the mainstream,so the research of this paper is based on similarity.This paper proposes three new link prediction algorithms by analyzing the existing methods based on similarity,to solve the problems at present.Now,the algorithms based on local topology similarity only consider the number of common neighbors and the degrees of common neighbor nodes between two nodes.If two pairs of nodes have the same number of common neighbors,and each common neighbor node has the same degree,then the similarity values of these two pairs of nodes are equal using algorithms based on local topology similarity,ignoring the fact that the topology is completely different between them.Therefore this paper proposes a new algorithm which considers the effect of edges between common neighbor nodes and uncommon neighbor nodes besides the degrees of common neighbor nodes.The existing algorithms based on local topology only consider the local topology information of networks,but ignore other social theories,such as community structure,strong and weak ties.Researchers have introduced community into link prediction algorithms,but these methods are defined under the premise of common neighbors,which think that the similarity value is zero between two nodes without common neighbors,so the prediction accuracy of these methods is not high.This paper uses community relevance to represent the similarity between different communities,and predict links by considering both local topology and community relevance.When two nodes are in different communities,this paper considers both local topology and community relevance,avoiding that the similarity values of too many node pairs are zero.Based on semi-local similarity,RALP algorithm has a better performance,but the algorithm does not take endpoints' own contribution to the similarity into account.This paper proposes a new semi-local similarity algorithm on the basis of RALP algorithm,considering endpoints' own contribution to the similarity.The algorithm not only considers the influence of different paths on similarity,but also punishes intermediate nodes which have larger degrees on the path with the same length,so the algorithm take endpoints' own contribution to the similarity into account.Through the simulations on the data sets of real social networks,this paper demonstrates that the proposed algorithms have higher prediction accuracy than the traditional link prediction algorithms based on similarity.