Multiple Rumor Source Detection with Graph Convolutional Networks

Quarantines as a Targeted Immunization Strategy

“…The study of epidemics on graphs is an active field of research. An important body of work assumes the underlying graph is known, and focuses on modeling epidemics [19,75,32,16,78,47], detecting whether there is an epidemic [8,9,54,53,51,44,40], finding communities [60,76], localizing the source of the spread [65,66,64,69,72,70,20] or instead obfuscating it [26,28,27], or controlling their spread [43,21,22,34,29,71,77,58]. The inverse problem, recovering the network from epidemic data, has also been extensively studied [56,1,18,59,41,35,33].…”

Learning Traffic Network Embeddings for Predicting Congestion Propagation

“…In addition to the traffic congestion propagation problems considered in this paper, our proposed methods for traffic congestion propagation can also be applied (with slight changes) to similar problems in other domains that require network embedding techniques. These include rumor spread (propagation) in social network [49], citation trend modeling and impact prediction in citation networks [50], user preference prediction via social network influence modeling (with applications of service recommendations in social media and online shopping websites) [51], etc. In these problems, the propagation of information (e.g., rumor, preference) poses similar propagation characteristics such as asymmetric transitivity, local proximity, and dynamic global tendency of propagations.…”

Schemes of Propagation Models and Source Estimators for Rumor Source Detection in Online Social Networks: A Short Survey of a Decade of Research

“…Nonetheless, they still suffer from the shortcoming that the node label is simply an integer which may restrict the prediction precision. To improve prediction precision on the same problem, Dong et al [10] then firstly attempts to apply the Graph Convolutional Networks (GCN) technique on multiple rumor source detection and then proposed a supervised learning based model named Graph Convolutional Networks based Source Identification (GCNSI). Basically, in GCNSI, Dong et al proposed an input generation algorithm to extend LPSI integer label into a multi-dimentional vector for each node as training data, and applied GCN in capturing different features of a node based on two assumptions, one is source prominence from LPSI, and the other is rumor centrality.…”