Acta Physica Sinica, Год журнала: 2025, Номер 74(12), С. 0 - 0

Опубликована: Янв. 1, 2025

Identifying key nodes in complex networks or evaluating the relative node importance with respect to others using quantitative methods is a fundamental issue network science. To address limitations of existing approaches—namely subjectivity assigning weights indicators and insufficient integration global local structural information—this paper proposes an entropy-weighted multi-channel convolutional neural framework (EMCNN). First, parameter-free entropy-based weight allocation model constructed dynamically assign multiple by computing their entropy values, thereby mitigating inherent traditional parameter-setting enhancing objectivity indicator fusion. Second, features are decoupled reconstructed into separate channels form feature maps, which significantly enhance representational capacity structure. Third, leveraging extraction capabilities power attention mechanisms, extracts deep representations from while emphasizing information through attention-based weighting, thus enabling more accurate identification characterization importance. validate effectiveness proposed method, extensive experiments conducted on nine real-world SIR spreading model, assessing performance terms correlation, accuracy, robustness. The Kendall correlation coefficient employed as primary evaluation metric measure consistency between predicted actual influence. Additionally, performed three representative synthetic further test model’s generalizability. Experimental results demonstrate that EMCNN consistently effectively evaluates influence under varying transmission rates, outperforms mainstream algorithms both accuracy. These findings highlight method’s strong generalization ability broad applicability tasks within networks.

Язык: Английский