Research On The Algorithm Of Identifying Influential Nodes Based On Complex Networks

With the rapid development of computer technology, complex network increasingly becomes an important research topic which is concerned by scholars at home and abroad. In the analysis and research of complex networks, researchers collected a large number of real data, and summarized the features of complex networks in different areas. Then, they found that the research of identifying influential nodes in complex network has a very important significance.In this dissertation, the topological characteristics of complex network structure are summarized and analyzed. The algorithms of identifying influential nodes in complex networks are studied further in the aspects of local structure and potential edge importance.Firstly, four models of complex network are introduced in this dissertation. By analyzing the statistical properties of complex network, it is observed that degree distribution, average path length and clustering coefficient play an important role in identifying influential nodes. Meanwhile the common algorithms used for the identification of influential nodes are summarized and these algorithms are compared each other.Secondly, aiming at the weighted network, the algorithm of identifying influential nodes based on evidence theory and local structure is proposed. In the algorithm, taking advantage of evidential centrality, the real degree distribution of each node in complex networks is considered. After that, combining the topological connections of the neighbors of a node, the influence value of the nodes can be obtained. The value the higher is, the influence of corresponding to the node the larger is. And then the influential nodes can be identified.Then, the algorithm of identifying influential nodes in unweighted network based on the potential edge importance is proposed. The potential importance of each edge is measured according to Jaccard similarity and an edge weighting method is proposed. Then evidential centrality to identify influential nodes is improved by taking the real degree distribution into consideration fully. Afterwards, k-shell decomposition method is used to measure the layer of nodes located in networks. And then the influence values of the nodes are achieved. These values are sorted by descending and the influential nodes can be identified.Finally, the experiments on some real datasets are done in MATLAB,and the performance of both of proposed methods in this dissertation is evaluated by doing comparison with several classical algorithms.