Study On Improved Algorithm Of Link Prediction

With the advancement of science and technology,more and more complex systems have emerged.Data derived from complex networks have increased straightly as well,which have in turn promoted the research process of complex networks.Link prediction is an important research direction of complex networks.It mainly utilizes the known data and their interactions to predict the data that already exists but has not been observed,the data that may appear in the future,and some of the false data.As the research results of link prediction are widely used in various fields,it is very important to improve the accuracy of link prediction.This dissertation focuses on the similarity between the nodes in the network,and mainly research the prediction methods based on the network topology.Firstly,this dissertation studies how to distinguish the different roles of common neighbors in unweighted and undirected networks by link strength and clustering coefficients of common neighbor nodes,and then proposes a new weighting method to study the link prediction problem in weighted networks.Finally,the problem of friend recommendation in social networks is investigated based on the PWCS phenomenon existing in the network.The main contents and contributions of the dissertation are as follows:1.In view of the fact that the existing link prediction algorithms could not well distinguish the different effects of common neighbor nodes on the formation of predicted links,this dissertation proposes the Node Link Strength and Clustering Coefficient link prediction index.The index combines with the link strength and clustering coefficient of the common neighbor nodes,and further distinguishes the common neighbor nodes.The experimental results on four real networks show that the NLSC index has obvious advantages compared with other similarity algorithms.2.Most of the weighted network link prediction methods only consider the natural weight of the link(the weight defined by the natural attribute of the link,such as the number of paper cooperation,flight frequency,etc.),ignoring the influence of the topological structure weight of the link on the prediction accuracy.A new weighted network method is proposed,which uses the clustering and diffusion characteristics of the edges in the network.By defining the topological structure weight and applies the weight to the existing weighted similarity index,a weighted prediction index based on link structure weight is proposed,including WCD-CN,WCD-AA,WCD-RA and WCD-LP.Experiments are performed on Matlab with AUC as the evaluation index.Simulation results show that the new WCD weighted prediction index has higher prediction accuracy.3.Considering the PWCS phenomenon and intermediary nodes are more inclined to introduce familiar people to target nodes,the WCDFR prediction index is proposed,which can well distinguish the relationship between candidate nodes and intermediary nodes.The experimental results on six real networks show that WCDFR index achieves a good prediction effect with a relatively low time complexity at the same time.