🤖 AI Summary
Community detection in multilayer networks is challenged by the coexistence of cross-layer consistency and intra-layer heterogeneity. Method: We propose a weighted layer-fusion spectral decomposition framework that unifies generalized eigenvector alignment with noise-robust regularization within a spectral model—departing from conventional single-layer or layer-averaging assumptions. Theoretically, it integrates Laplacian matrix perturbation analysis, multilayer graph signal processing, adaptive layer-weight learning, and stochastic block model derivation. Results: Evaluated on synthetic benchmarks and real-world academic multilayer networks (collaboration–citation–co-occurrence), our method achieves an average 12.7% improvement in F1-score over baselines including MSI and ML-Louvain. Its core contribution is the first unified spectral clustering framework for multilayer networks that simultaneously ensures theoretical interpretability and robustness to structural noise, thereby enhancing both accuracy and stability of community detection in complex relational systems.