Complementary bidirectional fusion for multi-view graph clustering

Multi-view graph methods emphasize integrating the information from different views to obtain promising performance. To achieve this goal, we introduce a novel model that performs Complementary Bidirectional Fusion with Cross-view Graph Filter(CBF-CGF). Specifically, CBF-CGF employs graph filters to capture high-order neighborhood information and learns a consistent graph embedding through an adaptive bidirectional fusion mechanism between similarity matrices and their spectral embeddings. This consistency graph embedding preserves both the global connectivity patterns captured by the similarity matrix and the local geometric structure inherited from the spectral embedding. Notably, adaptive weights are introduced to the indicator matrix to mine characteristics of different clusters. To tackle the optimization challenges, we propose an efficient iterative algorithm and provide a detailed analysis of its convergence and complexity. Extensive experiments validate that our method can yield comparable performance to state-of-the-art methods on nine real-world datasets, particularly on the COIL20 dataset where all evaluation metrics surpass 96.43 %.