부산대학교 도서관

Multi-view attribute graph clustering is a hot research topic in recent years. However, research on multi-view clustering is still relatively scarce, and all of them use graph convolution networks to process a single view, only obtaining insufficient feature representation. Since the data sources in real life are diverse, we propose a model that can handle multi-view data. Our proposed method can filter out high-frequency noise of views, process the features and correlations of multiple view nodes through attention networks, and form a comprehensive representation containing multi view information features and structural features. Then, by reconstructing the topology structure of each view through inner product, we ensure that the generated consistent representation contains the structural information of each view. Finally, we combine node embedding and clustering into one learning process through a soft clustering method. Experimental results have shown that our proposed model has better performance than the current state-of-the-art models.