A Deep Graph Network with Multiple Similarity for User Clustering in Human–Computer Interaction

User counterparts, such as user attributes in social networks or interests, are the keys to more natural Human–Computer Interaction (HCI). In addition, users’ and structures help us understand complex interactions HCI. Most previous studies have been based on supervised learning improve performance of However, real world, owing signal malfunctions devices, large amounts abnormal information, unlabeled data, unsupervised approaches (e.g., clustering method) mining particularly crucial. This paper focuses improving HCI proposes a deep graph embedding network with feature structure similarity (called DGENFS) cluster applications similarity. The DGENFS model consists Feature Graph Autoencoder (FGA) module, Structure Attention Network (SGAT) Dual Self-supervision (DSS) module. First, we design an attributed method divide users into clusters by making full use their attributes. To take advantage information human space, k-neighbor is generated between features. Then, FGA SGAT modules utilized extract representations features topological respectively. Next, attention mechanism further developed learn importance weights different effectively integrate structures. Finally, cluster-friendly features, DSS module unifies integrates learned from modules. explores high-confidence assignment soft label guide optimization entire network. Extensive experiments conducted five real-world data sets attribute clustering. experimental results demonstrate that proposed achieves most advanced compared nine competitive baselines.