Link Prediction Based On The Importance Of Network Nodes

There are various complex systems relate to daily life,such as transportation systems,biological systems and social relationships.All of these systems can be abstracted as networks.The entities of the system can be considered as the nodes of network and their connections are the links.Due to the complex relationships between these large numbers of entities,these related networks often exhibit complex network structures and called complex networks.Link prediction problem is a hot topic in the study of complex networks.It analyzes the possibility of links between unconnected nodes of the network in the future in order to mine more information of the network.Most link prediction algorithms measure the similarities between unconnected nodes to find potential links by using the structure information of the network which can be obtained easily.Taking into consideration of the community structure,node degrees and the importance of nodes,we propose some new link prediction algorithms for unweighted and weighted networks which improved the accuracy of link prediction.Based on community structures and node degrees,which are important topological structures of complex networks,we propose LAS similarity index describes the local structure of a pair of unconnected nodes and their common neighbors.Further,based on LAS index and considering the number of the common neighbors of a pair of unconnected nodes,we propose C-LAS index.Finally,considering different importance of the nodes in the network,three new link prediction algorithms are proposed,assigning a score to each common-neighbor nodes of a pair of unconnected nodes with degree centrality,betweenness centrality and closeness centrality,and distinguishing the different contributions of these nodes by adjusting the parameter ?.Five link prediction algorithms proposed in this thesis are applied to multiple unweighted networks to verify their effectiveness.Compared with the unweighted networks,weighted networks can represent real-world systems better.It is necessary to apply the importance of nodes into weighted networks.In the present thesis,three new algorithms,based on the parameter-dependent indices for WCN,are proposed for weighted networks,assigning a score to each common-neighbor nodes of a pair of unconnected nodes with the degree centrality,the betweenness centrality and the closeness centrality,and distinguishing the differentcontributions of these nodes by adjusting the parameter ?.The experiments of these algorithms on four weighted networks verify their effectiveness.