Community detection in attributed networks via adaptive deep nonnegative matrix factorization

Community detection plays an important role in analyzing attributed networks. It attempts to find the optimal cluster structures to identify valuable information. Although deep nonnegative matrix factorization (DNMF) is widely used in community detection, it cannot be used to analyze attributed networks since only topology information is considered. Recent researches have taken attribute information into account, but we still need to face the following challenges. First, it is difficult to deal with topology noise and attribute noise in attributed networks at one stroke. Second, we need to balance the coupling between topology and node attributes with hyperparameters in most methods. However, with inappropriate hyperparameters, it is easy to cause interference and compromise between them. For the above challenges, in this paper, we propose a novel method, namely adaptive deep nonnegative matrix factorization. Specifically, we handle the inherent noise of attributed networks via dual-DNMF with autoencoder. And then, we use the attention mechanism to adaptively integrate topology information and attribute information without adjusting hyperparameters manually. Overall, our method not only handles the inherent noise in attributed networks, but also resolves the interference and compromise between topology and attributes in a generalized way. The results of comprehensive experiments support our conclusions and demonstrate that our method outperforms the state-of-the-art methods in most datasets.