Deep Attention-guided Graph Clustering with Dual Self-supervision

Existing deep embedding clustering methods fail to sufficiently utilize the available off-the-shelf information from feature embeddings and cluster assignments, limiting their performance. To this end, we propose a novel method, namely attention-guided graph with dual self-supervision (DAGC). Specifically, DAGC first utilizes heterogeneity-wise fusion module adaptively integrate features of auto-encoder convolutional network in each layer then uses scale-wise dynamically concatenate multi-scale different layers. Such modules are capable learning an informative via attention-based mechanism. In addition, design distribution-wise that leverages assignments acquire results directly. better explore develop solution consisting soft strategy Kullback-Leibler divergence loss hard pseudo supervision loss. Extensive experiments on nine benchmark datasets validate our method consistently outperforms state-of-the-art methods. Especially, improves ARI by more than 10.29% over best baseline. The code will be publicly at https://github.com/ZhihaoPENG-CityU/DAGC.