Multi-view clustering integrating anchor attribute and structural information

Abstract

Multisource data has driven the development of advanced clustering algorithms, such as multi-view clustering, which critically rely on the construction of similarity matrices. Traditional algorithms typically generate these matrices based solely on node attributes. However, for certain directed real-world networks, neglecting the asymmetric structural relationships between nodes may compromise the accuracy of clustering results. This paper introduces a novel multi-view clustering algorithm, AAS, which employs a two-step proximity approach using anchors in each view, effectively integrating both attribute and directed structural information. This method enhances the clarity of cluster features within the similarity matrices. The construction of the anchor structural similarity matrix utilizes strongly connected components of directed graphs. The entire process—from the construction of similarity matrices to clustering—is formulated within a unified optimization framework. Comparative experiments conducted on the modified Attribute SBM dataset, benchmarked against seven other algorithms, demonstrate the effectiveness and superiority of AAS.