Attribute Disturbance for Attributed Multiplex Heterogeneous Network Embedding

Real-world networks are multiplex and possess a large number of attributes. With the advancement of embedding learning, many researchers have focused on Attribute Multiplex Heterogeneous Network Embedding. However, most existing studies have not fully leveraged node attributes and have not adequately captured the multiplex structure information. To address these challenges, we propose a method named Attributed Disturbance for Attribute Multiplex Heterogeneous Network Embedding(ADAMHE). ADAMHE deeply mines attribute features by enhancing node attributes and effectively integrates these attributes with relational structure information into the learning process of network embedding. Additionally, we utilize a self-attention mechanism to efficiently learn network embedding. To validate the efficiency and effectiveness of ADAMHE, we conduct comprehensive experiments on three real-world datasets, significantly outperforming other embedding baselines on all evaluation measures.