Key node recognition based on multidimensional feature extraction

Key node identification remains a prominent research focus within the domain of complex networks. Numerous researchers have proposed diverse methodologies predicated on structural characteristics, including network connectivity, distance metrics, eigenvalues and community structures. However, the identification of key nodes through the extraction of compressed network features remains a less explored avenue. To address this gap, this paper proposes a key node identification method predicated on the extraction of multi-dimensional network features. This approach involves extracting multi-dimensional features, specifically encompassing network visibility, phase space characteristics and state transition dynamics. Acknowledging the inherent nonlinear relationships among these features, a Random Forest algorithm is employed to model these interdependencies and perform the ultimate key node identification. The effectiveness of the proposed method was validated using the Susceptible-Infected-Recovered (SIR) model, with Kendall’s τ and Pearson correlation coefficients employed to assess ranking correlation. Preliminary investigations also assessed the impact of varying quantities of extracted features on the experimental results. The proposed method was evaluated on ten real-world network datasets and benchmarked against ten typical existing algorithms. Experimental results demonstrate that the proposed methodology offers significant advantages over extant approaches. This application of multidimensional feature extraction for identifying key nodes offers a novel perspective and a valuable analytical tool, thereby contributing to the advancement of research in this domain.