Research On Information Diffusion Source Localization On Social Networks

Due to the development of the Internet and the update of information technology,various social software has been interwoven to form intricate social network groups in recent years.These are the reasons why information of all kinds can spread rapidly in a very short time,causing uncontrollable public opinion.And the imperfection of today’s network supervision system makes the supervision of public opinion even more difficult.Society is unable to control the spread of negative information in a timely manner,causing far-reaching effects and countless losses of life and property.Therefore,how to track the source of information dissemination based on limited information in a shorter time has become a hot issue.Previous research methods mainly analyze the diffusion parameters of nodes in the network and their centrality for source localization,which mostly rely on path structure derivation and are computationally intensive,and the efficiency and accuracy need to be further improved.On this basis,this paper uses deep learning,path analysis and multilayer networks based on multiple propagation models to locate information sources,and the specific research work is as follows:Firstly,this paper proposes a deep learning-based algorithm for multi-information source localization.The algorithm considers path propagation and time series in information source diffusion.Based on this,a graph structured feature representation method is proposed.The algorithm uses a self-coding network for feature value extraction and ultimately obtains a source node score ranking to identify the propagation sources.Through experiments on two virtual networks and four real network datasets,the results show that the algorithm is able to locate the source nodes in less time based on less information.Secondly,this paper proposes an algorithm for locating information sources in time-varying networks based on link reconstruction.The algorithm starts from the propagation time-varying factors as well as the link recovery problem.By extracting the transformation laws through the correlation between physical connection and time,the proved method reconstructs the temporal network to determine the propagation path,and subsequently uses the infection subgraph to back-propagate the propagation path at the previous moment,so as to determine the location of the initial propagation source.As the experimental results in six temporal networks show,the algorithm achieves better performance further improving the accuracy of the information propagation source location problem.Thirdly,this paper proposes a multi-information source localization algorithm based on the social distance metric in dynamic networks.The algorithm considers the propagation patterns of multiple sources spreading in dynamic networks and adaptively adjusts them in conjunction with the hop of neighbors.Meanwhile,the proposed method is also combined with convolutional neural networks,which allows the accuracy of the algorithm to be improved.By conducting experiments in six dynamic networks,the results show that the algorithm is able to locate source nodes more accurately in a multi-source environment.