Research On Multi-Source Data Network Structure Inference And Balance Robustness Of Network

The era of complex networks has come with the development of technology and society.The flourishing development of complex networks has brought convenience to human society,whereas it also generates some negative effects.For example,some destructions or infectious diseases would spread more easily in the entire complex network system.In this case,the daily life of human beings is increasingly depended on the robustness and effectiveness of various complex network systems.Network robustness evaluation system is crucial for people to devise strategies,like the improvement of the robustness of networks,the mitigation of the destructive consequences under malicious attacks,and the defense of networks against all possible risks.Network structure inference is important to analyze various properties of networks(like the network robustness).This is because the link structure of a network reflects the potential functional relationship between the entities in the system.Generally,the real network structure of a system is first inferred,and then its network robustness would be analyzed and improved.Generally,link inference in unsigned networks aims to infer the possible unknown links between some nodes.For signed networks,their link structures usually are stable whereas their signs in links may be rapidly changed.The link inference in signed networks usually tries to determine the signs of the link structures.Actually,those link inference problems have the same properties when we model the signs “ +/-” of signed networks as the real/spurious links of unsigned networks.In this paper,we mainly study the link inference of networks and the analyses of the balance robustness of signed networks.The main works of this paper are illustrated as follows.(1)We propose a collective link inference model(CIM)to infer real and spurious links in networks based on the observed data from multiple platforms.In some applications,these observed data are unreliable,sparse and heterogeneous.The existing works are hard to infer the link structures of networks under those applications.To solve this problem,we study the link inference problem with sparse observation data and community structures.First,we aggregate those platforms into a set of clusters,in which platforms in the same cluster have similar observations and similar reliability.Second,we propose an expectation maximization algorithm(called C-EMLIC)to infer the link structure,considering the heterogeneous reliability of these platforms.Finally,experiments on the simulated networks and real networks show that C-EMLIC is superior to several classical algorithms in link inference when the observation data of platforms are sparse and the network has community structures.(2)A malicious attack on a complex network would result in catastrophic destructions on its potential functions.The research on the network robustness has received great attention in recent years.Previous research mainly focused on the impacts of malicious attacks on the robustness of unsigned networks.Compared with unsigned networks,signed networks with conflict relationships can represent more characteristics of real social systems,like the potential opposition and cooperation in complex networks.Although some progress has been made in the robustness of unsigned networks,the research on analyzing the robustness of signed networks is still in its infancy.Therefore,we study the effects of malicious attacks on the balance robustness of signed networks.First,we model two types of malicious attacks: small-scale node attacks and large-scale cluster attacks,and analyze the balance robustness of signed networks in the process of attacks.Second,we propose an index to evaluate the balance robustness of signed networks.Finally,we propose six protection strategies to protect some important nodes,so as to mitigate the damage of malicious attacks to the balance robustness of signed networks.The experimental results of both the synthetic signed networks and real-world signed networks show that the structural balance of signed networks is vulnerable to malicious target attacks,but its balance robustness can be improved significantly by protecting some important nodes.