Scalable Attributed-Graph Subspace Clustering

Over recent years, graph convolutional networks emerged as powerful node clustering methods and have set state of the art results for this task. In paper, we argue that some these are unnecessarily complex propose a model is more scalable while being effective. The proposed uses Laplacian smoothing to learn an initial representation before applying efficient self-expressive subspace procedure. This performed via learning factored coefficient matrix. These factors then embedded into new feature space in such way generate valid affinity matrix (symmetric non-negative) on which implicit spectral algorithm performed. Experiments several real-world attributed datasets demonstrate cost-effective nature our method with respect art.