Research On Influential Nodes Mining Algorithm In Complex Networks

The 21st century is a century of complexity science,with complex networks being the most typical.Benefiting from the rapid development of mobile Internet and Internet of things technology,network big data came into being and has grown rapidly.At present,the standards of node influence measurement are diversified,and the research on measurement methods are different.In our paper,we just focus on the method that based on local network topologies because of their low computational complexity.Since the accuracy of such methods is not optimistic currently,based on the centrality of the collective influence,we comprehensively utilize the local information such as the robustness,degree balance and inter-cluster connection strength of node's neighborhood,and propose two influence measurement methods that respectively corresponding for the two tasks of network transmission and virus control.Both of them can achieve good balance between the accuracy and time complexity.Then,we extend our focus from the node level to the network level to study the mining of influence node group.Considering that the accuracy of current heuristic algorithms is generally low,we innovatively propose an efficient node selection strategy based on the proposed influence measurement method,and similarly design two adjustable adaptive influence maximization algorithms based on local collective influence for different tasks.The main work is summarized as follows:(1)Combing the neighborhood robustness,degree balance and inter-cluster connection strength,we proposed a new method of collective influence centralization NewCI from the perspective of virus control.Based on the centrality of collective influence proposed for local tree-like network,we took the local neighborhood information of nodes into consdieration and designed NewCI.It can realize a more accurate measurement of node influence and has a stronger universality than the original method.Experimental results in artificial and real complex network datasets show that our approach and the betweenness centrality have the best performance compared to other centrality methods.In terms of both time complexity and execution efficiency,NewCI is superior to other centrality methods.(2)From the perspective of network transmission,we proposed NewCI~+,a collective influence centrality method based on neighborhood robustness.Based on the centrality of collective influence,we considered the neighborhood robustness and designed a more suitable measurement method NewCI~+in network transmission.By conducting experiments in real complex network datasets,we proved that it could perform as well as the closeness centrality.In terms of both time complexity and execution efficiency,it is also superior to other centrality methods.(3)Based on the proposed influence measurement methods NewCI~+,we designed the adjustable adaptive influence maximization algorithms LNewCI~+-ARIM under the selection strategy of local leader nodes.Considering that the collective influence centrality used by the CIM algorithm is not accurate enough in measuring the influence of nodes,the efficiency of nodes selection strategy is low,and there is a certain degree of propagation overlap,we improved it and innovatively proposed a new influence maximization algorithm LNewCI~+-ARIM for network transmission.It adopts the NewCI~+methods proposed in the third chapter to measure node influence and adjusts the selection strategy of the influence node compared with the CIM algorithm.Experiments on the propagation ability of multiple real datasets proved that our LNewCI~+-ARIM algorithms could realize a good balance between accuracy and time complexity.(4)Based on the proposed influence measurement methods NewCI,we designed the adjustable adaptive influence maximization algorithms LNewCI-ARIM under the selection strategy of local leader nodes.By replacing the influence measurement method in the CIM algorithm as NewCI,and adjusting the node selection strategy to be the same as(3),we proposed the influence maximization algorithm LNewCI-ARIM for virus control task and demonstrated its effectiveness through the network destructive experiment.