Year: 2026
Type: preprint
Abstract: Identifying influential nodes in directed complex networks is an important problem in network science. Although K-shell decomposition can efficiently identify network cores containing highly influential nodes, its low resolution and core-like group problems limit its effectiveness for key-node identification in directed networks. Many methods address these issues by integrating K-shell with additional features for node ranking, but they do not focus on improving shell ranking. In this paper, we propose a ranking method named \textit{Shell-based Structural-Propagation Influence} (SSPI) from a different perspective. First, a new directed K-shell representation, termed \textit{KSOutIn}, is introduced to consider both incoming and outgoing edges. The two components of KSOutIn are then analyzed to determine their ability to provide a reliable shell ranking. Since they are insufficient when used alone or through a weighted linear combination, structural and topological features are extracted and evaluated using Linear Regression, Random Forest, and Gradient Boosting Machine models across 16 real-world and synthetic directed complex networks. PFI and SHAP are then used to identify the most important features. Finally, one of the most important features is combined with the K-shell information to construct SSPI for node ranking. Evaluation on various real-world directed networks shows that SSPI achieves the highest overall accuracy with low imprecision and time complexity.
Source: Research Square
Author: Shima Esfandiari
Institution: Shiraz University
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225915294
