Research On Influential Nodes For Multi-relational Social Networks

The rise of online social media has led to the diversified nature of human social activities,and the entities involved in various social activities constitute a complex social system with sophisticated relation between them.Compared with the simple frameworks in network science like single-relational networks,multi-relational networks can represent more completely the diversified relation characteristics between nodes and has gradually become the mainstream of social network analysis.The heterogeneous characteristics expressed by entities often affect the structure and function of networks.Studying the influential nodes for multi-relational social networks can help to improve the scientific cognition level and efficient utilization capability for real social systems and has great practical significance for controlling the spread of the epidemic and guiding public opinion.However,the analytical frameworks of influential nodes developed so far are mainly devoted to single-relational networks.Systematic research achievements remain scarce for influential nodes in multi-relational networks.The modeling and representation of multi-relational social networks and the design of practical and efficient methods of influential nodes for the structural characteristics in multi-relational networks have become a challenging issue.Therefore,the thesis systematically studies related issues on influential nodes for multi-relational social networks by introducing analysis methods of multilayer networks and the multi-source information fusion technology based on the Dempster-Shafer evidence theory.The main research works are as follows:(1)The modeling and representation of multi-relational social networks are studied.On the basis of analyzing the advantages and disadvantages of existing research for multi-relational social network modeling methods and based on the multilayer network model and tensor representation framework,the abstraction and simplification are performed regarding the structural characteristics of multi-relational social networks,and a graph definition for multi-relational networks is studied and proposed;by adopting an “aligned-node type” multiplex network modeling,an equivalent representation framework for tensor representation and matrix representation is created from the basic definition of tensor and the diffusion dynamics equation for depicting multilayer networks.(2)The basic descriptors for multi-relational networks are studied using the analysis framework of multilayer networks.Based on the concept and resolving thoughts of the clustering coefficient for multilayer network,an elementary structure set of 3-cycles and a simplified calculating method are provided for the local clustering coefficient which is following the transitivity characteristics of multi-relational social networks.(3)For the issue of influential nodes in multi-relational undirected social networks,the impact of structure centrality and transitivity on influential nodes are studied with sociological theory,and the restrictive relationship between the basic descriptors and various influent factors is created.Ulteriorly,the thesis proposed multiplex ClusterRank(MCR)extending ClusterRank in single-relational networks to multi-relational networks.To address the problem of unsatisfactory performance of such method in large-scale networks,the correlation between the centrality and transitivity in multi-relational networks is studied,revealing the fact that due to the coupling relationship and the diversity of transfer mechanisms of multi-relational networks,there is no correlation between the centrality and transitivity.Dempster-Shafer evidence theory is further introduced and Multiplex ClusterRank is improved,thus studying and proposing Multiplex Evidential Centrality(MEC).(4)For multi-relational directed social networks,the difference of prestige and centrality regarding measuring directed relation influential nodes and the connection thereof are analyzed,and the IOMCR influential nodes sorting method is proposed based on IO-ClusterRank.For the problems and limitations of IOMCR,it is revealed that there is no correlation between the centrality,prestige and transitivity of multi-relational directed networks.Meanwhile,the Dempster-Shafer evidence theory is introduced and the measurement information for depicting centrality,prestige and transitivity is fused,thus further proposing the IOMEC.(5)For multi-relational directed weighting networks,on the basis of analyzing the effect of relation strength on influential nodes and based on the number of behavioral interactions between nodes,a quantitative method for relation intimacy is proposed and used as a quantitative index for depicting relation strength;based on IOMEC,an evidential centrality method fusing strength information-Multiplex Strength Evidential Centrality(MSEC)is proposed and used for influential nodes analysis in multi-relational directed weighting social networks;(6)A great number of experiments are carried out on multiple real network data sets.The proposed sorting method is evaluated from two aspects: robustness and vulnerability,transmission dynamics characteristics.While verifying the validity of each method,the advantage and necessity of the thought of influential nodes sorting with multi-information fusion are demonstrated.Numerous experiments were carried out on various real network data sets,and the proposed ranking methods were evaluated in terms of robustness,vulnerability,and dynamics.The experimental results verified the three proposed methods(MEC,IOMEC,MSEC)can effectively eliminate the influence of the coupling information and the transmission mechanism in multi-relational networks,can more accurately identify influential nodes in the network,further demonstrate the advantage and necessity of the idea of influential nodes sorting with multi-information fusing multi-source information.The thesis work enriches the analysis framework for multi-relational networks,provides new ideas and methods for identifying influential nodes in multi-relational social networks,and further expands the application of information fusion technology.