CM-GCN: Crossbred Method based-Graph Convolution Networks for identifying influential spreaders from directed networks

• Main Authors: Saha, Nilanjana, Namtirtha, Amrita, Dutta, Animesh
• Format: Conference Proceeding
• Language: English
• Subjects: Convolution; Convolutional neural networks; Correlation; Crossbred method; Directed Network; DSIR epidemic model; Epidemics; Graph Convolution Network; Influential Spreaders; Kendall τ rank correlation; Task analysis
• ISSN: 2155-2509
• DOI: 10.1109/COMSNETS59351.2024.10427097

Influential spreaders play a critical role, either maximizing information dissemination or controlling epidemic spreads. Much of the existing research concentrates on identifying optimal spreaders in undirected networks. However, recognizing the significance of edge direction in spreading processes is essential when pinpointing influential spreaders. Graph Convolutional Neural Networks (GCNs) have shown remarkable efficacy in various graph-based tasks. Nevertheless, the challenge of over-smoothing in GCNs leads to nodes converging to similar representations with increasing depth, restricting the practical usage of shallow models. In response, this paper introduces a novel approach, the Crossbred Method-based Graph Convolutional Network (CM-GCN), to identify superior spreaders, considering the intricacies of directed networks. The proposed CM-GCN method considers the spreading dynamics specific to directed networks. Through a series of experiments on real-world datasets, our results demonstrate that the CM-GCN model outperforms prevailing baseline approaches. This advancement is particularly unable compared to crossbred method and other centrality methods in directed networks. Moreover, the CM-GCN method effectively addresses over-smoothing in deep GCNs while mitigating information loss attributable to DropEdge. The spreading performance of the proposed method, evaluated using the Directed Susceptible-Infected-Recovered (DBIR) epidemic model on six real networks, showcases significant enhancements in spreading dynamics compared to the established crossbred method.