Research On Identification Of Node Influence In Signed Networks

Identifying influential nodes in networks has important social value and practical significance.Most traditional influential-node-identification methods are based on unsigned networks which only consider positive relationships between individuals.However,in a real network environment,there are not only positive but also negative links between nodes.Although some scholarly works have been carried out on ranking the nodes in a signed network simply adapting traditional algorithms,the positive and negative links between nodes are not fully considered.To accurately identify the influential nodes in signed networks,this work provides an influence-calculation method based on the dependence relationship among signednetwork nodes.The power-dependence influence algorithm,denoted PD-Inf,measures the influence brought to the target node by the power-dependence relationship between nodes.The main work in this thesis is divided into two parts,an introduction of PD-Inf technique specifically designed for signed networks and an illustration of the concept.Firstly,we consider the inequality of dependences due to the differences in structural properties of nodes based on Emerson’s exploration of power-dependence theory and quantify the differences in the degree of dependence to analyze node influence.Furthermore,we also discuss how a triangle structure affects the dependence relationships based on the structural balance theory and propose an algorithm for identifying influential nodes in signed networks.Secondly,we illustrate the proposed algorithm by verifying it on real signed social network datasets.The experimental results demonstrate that PD-Inf can evaluate the influence of a node in signed networks effectively by considering both positive and negative links.In addition,an application of the proposed algorithm in an unsigned network is also studied and analyzed.The influence ranking of nodes determined by PD-Inf can be divided into three groups.A small topranked group of influential users can cause the information to propagate more widely in the network under certain diffusion conditions,and the influence coverage of such nodes is larger;the PD-Inf has a poor influence distinction of mid-ranked users;bottom-ranked users can spread information quickly thus can be marked as fast nodes in the network.