Research On Rumor Source Detection And Location In Social Networks

Research on the Detection and Localization Methods of Rumor Sources is an important component of the field of social network analysis.Source localization on social networks involves locating the source node of a propagation entity after it has spread on the network,determining the location of the source of propagation and inferring its initial propagation time.These issues have long been of concern to experts and scholars.In real life,some rumors often spread through social networks,posing a serious threat to the public property and safety,and causing very adverse consequences.However,the sources of these rumors are often only a few nodes,but through the complex structure of social networks,they transmit negative information to the vast majority of people,leading to large-scale rumor dissemination.Therefore,in order to curb the adverse effects of rumors and minimize the harm they bring,it is crucial to find the source node of the rumors’ initial propagation.This process is the detection and localization of rumor sources.However,rumors are all spread on the network.Due to the huge scale of today’s networks and the complex links between nodes,the propagation of rumors in actual networks is highly random.Most existing rumor source localization methods are based on the IC model’s maximum likelihood estimation,which is inefficient and does not fully utilize the initial propagation time of the rumor source.Therefore,how to quickly and accurately detect and locate the source of rumors based on time information is a challenging research topic.This research has significant implications and broad practical application prospects and value.Regarding the research on detecting and locating the source of rumors in social networks,the main work of this paper includes:(1)We proposed a TRU model(Truth-Rumor-Unknown)for multi-entity rumor propagation with the existence of debunking information,and based on this,we propose an influence source detection and localization method based on maximum a posteriori estimation.In the real world,many rumors are spread competitively with true information.To address this,we propose a more practical rumor spreading threshold model(TRU model)that considers different nodes having different degrees of credibility when spreading information.Based on this,we propose a maximum a posteriori method to rank candidate seed nodes and determine the source node.Experiments on three virtual networks and five real network datasets show that our algorithm can more accurately locate the source node under more realistic conditions.(2)We proposed a influence source localization method based on Markov Chain Monte Carlo.This method integrates time information and uses Markov Chain Monte Carlo method to perform extensive sampling on a specified distribution network,and draws propagation delay samples.First,we find the shortest propagation path length as the transmission delay between two points.Secondly,we use the correlation coefficient to map the correlation between the propagation delay sample and the observed node information.Finally,the candidate seed nodes are ranked according to the correlation coefficient,and the node with the highest correlation coefficient is selected as the propagation source.Through experiments on three virtual networks and five real networks,the results show that the algorithm has good performance and further improves the accuracy of the rumor propagation source localization problem.(3)We proposed a joint multi-channel influence source localization method.This method considers the competitive dissemination of different information and the different credibility of nodes based on the TRU threshold model.It normalizes SI-TRU,betweenness centrality,and closeness centrality and then performs a linear combination.Different combinations of parameters are tested,and the most stable parameters with the highest source localization accuracy are selected for use in source localization.Finally,the node with the maximum combined value is selected as the source node.Experimental results on three virtual networks and five real networks show that this algorithm can more accurately locate the source node in the environment where true and false information is disseminated together..