Identifying Multiple Propagation Sources With Motif-Based Graph Convolutional Networks for Social Networks

Identifying the sources of propagation in social networks, such as misinformation propagation, is one key issues recently. Most existing studies assume underlying model known, which difficult to obtain practice. Recent efforts have been devoted detect multiple real-world situations, and influence neighbors assumed be identical. However, this assumption will result inaccurate results infection state a node determined by its critical neighbors. In paper, we fill gap capturing with structural properties networks. For instance, opinions are more likely spread via closely connected friends within small groups. Here propose Motif-based Graph Convolutional Networks for Source Identification (MGCNSI) framework based on GCN-based source identification approach. Specifically, different network motifs used capture types properties. Then each motif extracts particular type, motif-based graph convolutional layer constructed aggregate that motif. To adapt mechanisms, an attention mechanism aggregation designed automatically assign higher weights informative motifs. The empirical demonstrate MGCNSI outperforms several benchmark methods both synthetic advantage most obvious networks denser neighborhoods, where can select from larger neighbor sets. How paths also illustrated.