<p>Processing large-scale attributed networks with millions of nodes and billions of edges requires efficient high-performance computing methods to handle the computational complexity of integrating both structural and attribute information. Community detection in node-attributed networks, where nodes are characterized by both structural connections and attribute information, is a crucial task in network analysis. Accurately identifying these communities can reveal underlying patterns and relationships within the network, providing deeper insights into its structure and behavior. However, the challenge lies in effectively integrating both structural and attribute data to detect these communities. We propose a novel method, consensus-based dual-spectral embedding (CDSE) for attributed graph clustering, to address this challenge. Our approach calculates cosine similarity in the attribute space and Jaccard similarity in the structure space, selecting eigenvectors from each. These eigenvectors are then combined using a consensus-based approach to create an embedding space vector that ensures balanced representation from both spaces. This unified representation enhances community detection by reflecting both structural and attribute similarities. The algorithm’s O(<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(n^2K\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mi>n</mi> <mn>2</mn> </msup> <mi>K</mi> </mrow> </math></EquationSource> </InlineEquation>) computational complexity and modular parallel design make it suitable for deployment on high-performance computing systems. Extensive experiments on real-world datasets demonstrate that our method outperforms traditional algorithms in accuracy and robustness. The results highlight the effectiveness and potential of CDSE in identifying meaningful communities in node-attributed networks, with scalability for large-scale graph processing in HPC environments.</p>