Key nodes identification in complex networks based on subnetwork feature extraction
The problem of detecting key nodes in a network (i.e. nodes with the greatest ability to spread an infection) has been studied extensively in the past. Some approaches to key node detection compute node centrality, but there is no formal proof that central nodes also have the greatest spreading capa...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2023-07-01
|
| Series: | Journal of King Saud University: Computer and Information Sciences |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1319157823001854 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The problem of detecting key nodes in a network (i.e. nodes with the greatest ability to spread an infection) has been studied extensively in the past. Some approaches to key node detection compute node centrality, but there is no formal proof that central nodes also have the greatest spreading capacity. Other methods use epidemiological models (e.g., the SIR model) to describe the spread of an infection and rely on numerical simulations to find out key nodes; these methods are highly accurate but computationally expensive. To efficiently but accurately detect key nodes, we propose a novel deep learning method called Rank by Graph Convolutional Network, RGCN. Our method constructs a subnetwork around each node to estimate its spreading power; then RGCN applies a graph convolutional network to each subnetwork and the adjacency matrix of the network to learn node embeddings. Finally, a neural network is applied to the node embeddings to detect key nodes. Our RGCN method outperforms state-of-the-art approaches such as RCNN and MRCNN by 11.84% and 13.99%, respectively, when we compare the Kendall’s τ coefficient between the node ranking produced by each method with the true ranking obtained by SIR simulations. |
|---|---|
| ISSN: | 1319-1578 |