A key node identification method in CSI 300 network based on node centrality and community information
Аннотация
CSI 300 stock index consists of the 300 largest and most liquid securities in the Chinese stock market. Transforming stock data into complex networks enables the identification of key companies that influence market volatility and systemic risk propagation. Existing studies on key node identification in CSI 300 networks mainly rely on single statistical indicators, which are insufficient to fully capture the structural heterogeneity of nodes and their interaction patterns. To address this limitation, this paper proposes a novel centrality framework termed Leiden-Principal Component Analysis (L-Pca). The method integrates Leiden community detection with multi-dimensional structural feature fusion. Specifically, node feature vectors are constructed by combining global centrality measures, community-relative centrality, and neighborhood structural attributes. Principal component analysis is subsequently applied to perform feature fusion and dimensionality compression, generating composite node importance scores in an orthogonal structural space. To evaluate the effectiveness and robustness of the proposed method, extensive experiments are conducted on six representative networks using connectivity-based evaluation, ranking monotonicity analysis, and SIR-based spreading simulations. Moreover, robustness analyses under node perturbations and different network filtering strategies are performed to verify the stability of the identified influential nodes. Experimental results demonstrate that L-Pca achieves superior performance in key node identification compared with most baseline centrality methods, while maintaining strong robustness and interpretability.